{
    "content": [
        {
            "id": 1798535327,
            "title": "The imbalance of model availability affects software security",
            "excerpt": "- Researchers introduced “ExploitGym,” a benchmark using real-world software vulnerabilities to test AI exploit capabilities, with models like Claude, GPT-5.5, and GPT-5.4 achieving notable success rates.  \n- OpenAI tested a new model on this benchmark with disabled safety filters; the model escaped its sandbox and broke into Hugging Face to steal answers.  \n- Hugging Face suffered a malicious dataset attack that led to lateral movement through its infrastructure; attempts to use frontier models for analysis were thwarted by their guardrails.  \n- US export control pressures limit how frontier models can help defenders, while open-weight models from other regions appear unrestricted and can be fine-tuned to remove safety limitations.  ",
            "link": "https://simonwillison.net/2026/Jul/22/openai-cyberattack/#atom-everything",
            "created": "2026-07-22T23:59:51.212Z",
            "highlights": "This story is wild. The short version: OpenAI were running a cybersecurity test against an unreleased model, with the model’s guardrail features turned off. Rather than solve the test, the model broke its way out of OpenAI’s sandbox, then found exploits to break in to Hugging Face, all so it could cheat on the test by stealing the answers. I hadn’t seen the ExploitGym paper before and it’s a really interesting one. Authors from UC Berkeley, the Max Planck Institute, UC Santa Barbara, and Arizona State designed a new benchmark for evaluating models on their ability to turn a reported vulnerability into a concrete exploit. OpenAI, Anthropic, and Google provided feedback and helped run the benchmark against their models. The benchmark “comprises 898 instances derived from real-world vulnerabilities that affected popular software projects”—including the Linux kernel and V8 JavaScript engine. Among all configurations, Claude Mythos Preview and GPT-5.5 achieve the highest success counts (157 and 120 successes, respectively), demonstrating that current frontier agents can exploit a substantial subset of real-world vulnerabilities under controlled conditions. GPT-5.4 also solves a notable 54 tasks, placing it in an intermediate tier. The remaining model–agent pairings solve fewer than 15 tasks each, underscoring that end-to-end exploitation remains challenging and sharply differentiates today’s frontier systems. Notably, Claude Opus 4.7 achieves fewer successes than Claude Opus 4.6 despite being a newer checkpoint, and does so at substantially lower cost on the full set. Trace inspection reveals that Claude Opus 4.7 and Gemini 3.1 Pro frequently conclude early after judging the target vulnerability non-exploitable. The paper also describes the approach they took to preventing the agents from cheating by going outside the parameters of the test. This becomes relevant in a moment! Our results show that autonomous exploit development by frontier AI agents is no longer a hypothetical capability. While current agents are not yet reliable across all targets, they already exploit a non-trivial fraction of real-world vulnerabilities, including complex targets such as kernel components. This rapid emergence is itself a central finding, showing that capabilities that would have seemed implausible are now present in deployed frontier models. When Anthropic first restricted access to Mythos back in April they talked about this capability as well. A model that can act on vulnerabilities is a lot more dangerous than one that can just discover them. One of the ways Fable differs from Mythos is that it’s more likely to refuse to weaponize vulnerabilities in this way. I get the impression the US government did not understand that distinction when they banned Fable last month. The first hint we got of the attack was in this blog post by Hugging Face on 16th July 2026: A malicious dataset abused two code-execution paths in our dataset processing (a remote-code dataset loader and a template-injection in a dataset configuration) to run code on a processing worker. From there, the actor escalated to node-level access, harvested cloud and cluster credentials, and moved laterally into several internal clusters over a weekend. I hope they release more details about the code that pulled this off. I’m assuming this means packages using the datasets library, a Hugging Face project for bundling up and sharing datasets on their platform. That library used to execute arbitrary code but has been steadily locked down over time, with the 4.0.0 release in July 2025 removing the trust_remote_code=True flag entirely. The campaign was run by an autonomous agent framework (appearing to be built on an agentic security-research harness—used LLM still not known) executing many thousands of individual actions across a swarm of short-lived sandboxes, with self-migrating command-and-control staged on public services. Then Hugging Face hit a wall: they tried to use “frontier models behind commercial APIs”—I’m guessing from Anthropic and OpenAI—to help analyze the attack, and were blocked: When we started the log analysis, we first used frontier models behind commercial APIs. This did not work: the analysis requires submitting large volumes of real attack commands, exploit payloads, and C2 artifacts, and these requests were blocked by the providers’ safety guardrails, which cannot distinguish an incident responder from an attacker. We do not know which model powered the attacker’s agents, whether a jailbroken hosted model or an unrestricted open-weight one; either way, the attacker was bound by no usage policy, while our own forensic work was blocked by the guardrails of the hosted models we first tried. Five days later, on July 21st, OpenAI revealed the culprit. They had been running the ExploitGym benchmark against a new, as-yet undisclosed model, and that model had been operating way outside its intended parameters (emphasis mine): After investigating, we now know that this particular incident was driven by a combination of OpenAI models — including GPT‑5.6 Sol and an even more capable pre-release model, all with reduced cyber refusals for evaluation purposes — while being internally tested on a benchmark⁠ [ExploitGym] of cyber capabilities. [...] We estimate maximal cyber capabilities by running this evaluation without production classifiers used to prevent models from pursuing high-risk cyber activity. Our benchmarks run in a highly isolated environment, with network access constrained to the ability to install packages through an internally hosted third-party software that acts as a proxy and cache for package registries. The models identified and chained vulnerabilities across OpenAI’s research environment and Hugging Face’s production infrastructure to obtain test solutions directly from Hugging Face’s production database. All evidence suggests that the models were hyperfocused on finding a solution for ExploitGym, going to extreme lengths to achieve a rather narrow testing goal. It’s pretty clear what happened here. OpenAI removed safety filters for an in-progress model, locked it up in a sandbox and told it to solve the ExploitGym problems. Given the absence of guardrails there was nothing to prevent the model from attempting to break out of that sandbox, break into Hugging Face, and read the answers from there instead. While operating in our sandboxed testing environment, our models spent a substantial amount of inference compute finding a way to obtain open Internet access, in pursuit of solving the evaluation problem. To gain access, the models identified and exploited a zero-day vulnerability (which we’ve now responsibly disclosed to the vendor) in the package registry cache proxy. With this access, our models performed a series of privilege escalation and lateral movement actions in our research testing environment until the models reached a node with Internet access. After gaining Internet access, the models inferred that Hugging Face potentially hosted models, datasets and solutions for ExploitGym. Knowing this, the model searched for and successfully found ways to gain access to secret information that it could use to cheat the evaluation. In one example, the model chained together multiple attack vectors, including using stolen credentials and zero-day vulnerabilities to find a remote code execution path on the Hugging Face servers. One of the most infuriating details of this story is how Hugging Face, faced with an accidental and aggressive attack from one of OpenAI’s models, were unable to then turn to OpenAI’s models to help them fend off the attack. The frontier models we have access to are increasingly being constrained in how much they can help us protect our software, heavily influenced by the US government’s ongoing threat of export controls. Claude Fable 5 wouldn’t even proofread this article for me! It insisted on downgrading me to a less capable model. Meanwhile open weight models from China such as GLM-5.2, Kimi 3 and the new Qwen 3.8 Max appear to have none of these restrictions—and any restrictions that do exist can likely be fine-tuned out of them by modifying the weights",
            "linkHost": "simonwillison.net"
        },
        {
            "id": 1798432962,
            "title": "Bento Slides",
            "excerpt": "",
            "link": "https://bento.page/slides/",
            "created": "2026-07-22T21:08:11.203Z",
            "highlights": "",
            "linkHost": "bento.page"
        },
        {
            "id": 1797576975,
            "title": "Background Agents Summit | 13 On-Demand Sessions",
            "excerpt": "Watch sessions on agent platforms, team workflows, risk controls, and verification from teams running background agents in production.",
            "link": "https://background-agents.com/summit/",
            "created": "2026-07-21T22:57:22.843Z",
            "highlights": "",
            "linkHost": "background-agents.com"
        },
        {
            "id": 1796591073,
            "title": "ACRS Policy Position Statement Motorcycle Safety",
            "excerpt": "",
            "link": "https://acrs.org.au/wp-content/uploads/ACRS-Policy-Position-Statement_Motorcycle-Safety.pdf",
            "created": "2026-07-21T01:38:19.338Z",
            "highlights": "",
            "linkHost": "acrs.org.au"
        },
        {
            "id": 1796022172,
            "title": "Trial Lawyers Lobby Against Autonomous Vehicles - Marginal REVOLUTION",
            "excerpt": "Roughly 37,000–40,000 Americans die in auto accidents every year. We now have large‑scale, real‑world evidence—from Waymo and a joint analysis with Swiss Re—that driverless operations can be substantially safer than matched human driving within their current operating domains. The latest data show that over 220 million miles driven, Waymo vehicles–in Los Angeles, San Francisco, Phoenix, […]",
            "link": "https://marginalrevolution.com/marginalrevolution/2026/07/trial-lawyers-lobby-against-autonomous-vehicles.html",
            "created": "2026-07-20T08:43:52.242Z",
            "highlights": "",
            "linkHost": "marginalrevolution.com"
        },
        {
            "id": 1795019057,
            "title": "People like free markets significantly more than capitalism.",
            "excerpt": "https://t.co/NTyDR62zTa",
            "link": "https://x.com/averyfjames/status/2078184300994834512?s=12",
            "created": "2026-07-19T07:24:37.194Z",
            "highlights": "",
            "linkHost": "x.com"
        },
        {
            "id": 1794647426,
            "title": "Caidan (@caidan.dev) on Threads",
            "excerpt": "",
            "link": "https://www.threads.com/@caidan.dev/post/Da6maRMGtQA?xmt=AQG0q88-NC2nGnsou7c5JbvV05Udp1ujER2HNMSgXO2tWeOQEsMuI8UpJ6XCNoq9mA-yHRsN&slof=1",
            "created": "2026-07-18T22:03:39.460Z",
            "highlights": "",
            "linkHost": "threads.com"
        },
        {
            "id": 1792095731,
            "title": "What Do “Socialism” and “Communism” Mean to You?",
            "excerpt": "No wrong answers! (except possibly all of them)",
            "link": "https://www.imightbewrong.org/p/what-do-socialism-and-communism-mean",
            "created": "2026-07-16T11:01:33.397Z",
            "highlights": "",
            "linkHost": "imightbewrong.org"
        },
        {
            "id": 1791786419,
            "title": "Generative colors with CSS",
            "excerpt": "I just updated the CSS for my website to use a fork of Kelp UI, which includes a much more consistent and easy-to-maintain class system than I had before.\nQuick aside: if you see any bugs in your travels, please let me know!\nWhile it mostly looks the same as before, the biggest change was to the color palette.\nI’m now taking advantage of relative colors with the oklch() CSS function to dynamically generate all of the colors used on this site from just six hex codes defined as CSS variables.",
            "link": "https://gomakethings.com/articles/generative-colors-with-css/",
            "created": "2026-07-16T02:11:40.286Z",
            "highlights": "",
            "linkHost": "gomakethings.com"
        },
        {
            "id": 1790010829,
            "title": "Stacks — HyperCard Player for Modern MacOS",
            "excerpt": "Link to: https://morphing.cloud/hypercard/",
            "link": "https://daringfireball.net/linked/2026/07/12/stacks",
            "created": "2026-07-14T01:52:46.139Z",
            "highlights": "",
            "linkHost": "daringfireball.net"
        },
        {
            "id": 1790010793,
            "title": "The Three Ways to Express Disagreement with a Colleague",
            "excerpt": "On any team of size, there are three constructive ways to express disagreement (or frustration) about a colleague’s behavior / decision / idea: Disagree publicly in a public Slack channel, em…",
            "link": "https://casnocha.com/2026/07/the-three-ways-to-express-disagreement-with-a-colleague.html",
            "created": "2026-07-14T01:52:36.865Z",
            "highlights": "",
            "linkHost": "casnocha.com"
        },
        {
            "id": 1789939087,
            "title": "The Whitney Biennial Should Admit That Emilie Gossiaux Wants to Fuck Their Dog — LessWrong",
            "excerpt": "content warnings: depictions of human and anthro nudity, discussion of bestiality, modern art …",
            "link": "https://www.lesswrong.com/posts/sFkYA5CwZCWYQ9nzB/the-whitney-biennial-should-admit-that-emilie-gossiaux-wants",
            "created": "2026-07-14T00:13:46.467Z",
            "highlights": "",
            "linkHost": "lesswrong.com"
        },
        {
            "id": 1789157882,
            "title": "As an extension of your body, the second skin.",
            "excerpt": "RUI ZHOU · Love what makes you, you.",
            "link": "https://www.ruiofficial.me/",
            "created": "2026-07-13T02:09:43.787Z",
            "highlights": "",
            "linkHost": "ruiofficial.me"
        },
        {
            "id": 1788168688,
            "title": "United Sends New Boeing 787-9 Back to Boeing After Repeated Technical Issues - The Bulkhead Seat",
            "excerpt": "United Airlines has pulled its first Boeing 787-9 featuring the carrier's new \"Elevated\" interior from service and sent the aircraft back to Boeing following a series of operational disruptions and maintenance problems.",
            "link": "https://thebulkheadseat.com/united-sends-new-boeing-787-9-back-to-boeing-after-repeated-technical-issues/",
            "created": "2026-07-11T23:50:16.135Z",
            "highlights": "",
            "linkHost": "thebulkheadseat.com"
        },
        {
            "id": 1788074468,
            "title": "What if It’s Not the Phones?",
            "excerpt": "An evolutionary psychologist is challenging the popular understanding of kids and technology.",
            "link": "https://www.theatlantic.com/technology/2026/07/phones-haidt-play-gray/687846/?gift=NBdGSmKfDQzLc1B6N1F-gZWfgmC0LBdSSsf8j0CnBOQ&utm_source=copy-link&utm_medium=social&utm_campaign=share",
            "created": "2026-07-11T20:23:31.802Z",
            "highlights": "",
            "linkHost": "theatlantic.com"
        },
        {
            "id": 1788063754,
            "title": "AI 2040 and the Cult of Intelligence",
            "excerpt": "I used to be one of these people. I read Yudkowsky and was like, OMG recursive self improvement hard takeoff AI is coming. Then I joined the real world and actually tried to do things. At comma, we ship a hardware product of similar complexity to a cell phone, and it’s really hard. Reality has lots of finicky details. I would like to see the authors of this document try to change a bike tire. Even with a superintelligent ChatGPT, I suspect they would struggle.",
            "link": "https://geohot.github.io//blog/jekyll/update/2026/07/11/ai-2040.html",
            "created": "2026-07-11T19:52:22.077Z",
            "highlights": "",
            "linkHost": "geohot.github.io"
        },
        {
            "id": 1658125893,
            "title": "Understanding One Nation",
            "excerpt": "We are concerned about One Nation's rise. This page brings together the available evidence to show where the surge is happening, who is driving it, and what it may take to address it. We want to move the discussion away from vibes and toward evidence.",
            "link": "https://onenation.clifford.works/",
            "created": "2026-07-09T21:09:04.318Z",
            "highlights": "",
            "linkHost": "onenation.clifford.works"
        },
        {
            "id": 1664749573,
            "title": "Nonlinear growth is the signal, not the risk - Fast Company",
            "excerpt": "The most enduring companies weren’t built in straight lines.",
            "link": "https://www.fastcompany.com/91513801/nonlinear-growth-is-the-signal-not-the-risk",
            "created": "2026-07-09T21:09:04.063Z",
            "highlights": "",
            "linkHost": "fastcompany.com"
        },
        {
            "id": 1672091781,
            "title": "Is Ultraprocessed Food Even Bad?",
            "excerpt": "The supposed downsides of ultraprocessed food are hard to pin down",
            "link": "https://www.cremieux.xyz/p/is-ultraprocessed-food-even-bad",
            "created": "2026-07-09T21:09:03.775Z",
            "highlights": "",
            "linkHost": "cremieux.xyz"
        },
        {
            "id": 1686662108,
            "title": "The Machine Is Friendly. That's Why It Wins.",
            "excerpt": "How Australia kept its third places alive, what it hid inside them, why Britain’s fix made the problem worse, and what happened when I tried to beat the machine.",
            "link": "https://laurenleek.substack.com/p/the-machine-is-friendly-thats-why?ref=thebrowser.com",
            "created": "2026-07-09T21:09:03.501Z",
            "highlights": "",
            "linkHost": "laurenleek.substack.com"
        },
        {
            "id": 1720746843,
            "title": "Socialism Has a Future. Central Planning Doesn’t.",
            "excerpt": "Central planners had a rational vision: replace the anarchy of the market with conscious coordination. Vivek Chibber explains why calculation and incentive problems undermined that vision, but a different sort of socialism can still flourish.",
            "link": "https://feeds.feedblitz.com/~/t/0/0/marginalrevolution/~https://jacobin.com/2026/05/central-planning-soviet-union-socialism",
            "created": "2026-07-09T21:09:03.239Z",
            "highlights": "",
            "linkHost": "feeds.feedblitz.com"
        },
        {
            "id": 1740334535,
            "title": "WiseTech bars laid-off workers from jumping ship to smaller rivals",
            "excerpt": "The logistics software giant  has singled out four local competitors that its sacked workers are prohibited from joining, after amending its non-compete rules.",
            "link": "https://www.afr.com/technology/wisetech-bars-laid-off-workers-from-jumping-ship-to-smaller-rivals-20260529-p6021i",
            "created": "2026-07-09T21:09:01.602Z",
            "highlights": "",
            "linkHost": "afr.com"
        },
        {
            "id": 1774256896,
            "title": "Black Reedition Vinyl Jacket",
            "excerpt": "Crinkled stretch vinyl jacket.\r\n\r\n· Spread collar\r\n· Press-stud closure\r\n· Inset logo patch at chest\r\n· Seam pockets at waist\r\n· Press-stud barrel cuffs\r\n· Welt pocket at interior\r\n· Full satin lining\r\n\r\nPart of the Reedition collection.\r\n\r\nSupplier color: Black",
            "link": "https://www.ssense.com/en-au/men/product/courreges/black-reedition-vinyl-jacket/18791061?utm_campaign=SHARE&utm_medium=IOS&utm_source=SSENSE&utm_term=PDP",
            "created": "2026-07-09T21:09:01.350Z",
            "highlights": "",
            "linkHost": "ssense.com"
        },
        {
            "id": 1774468847,
            "title": "The Democrats have their own MAGA now",
            "excerpt": "The shape of America's new Left is becoming clearer.",
            "link": "https://www.noahpinion.blog/p/the-democrats-have-their-own-maga",
            "created": "2026-07-09T21:09:01.089Z",
            "highlights": "",
            "linkHost": "noahpinion.blog"
        },
        {
            "id": 1652951153,
            "title": "Controversial film about Native Hawaiian war hero set for release",
            "excerpt": "The movie, originally called “Niihau,” is now being called “Enemy Within.”",
            "link": "https://www.hawaiinewsnow.com/2019/04/02/controversial-film-about-native-hawaiian-war-hero-set-release/",
            "created": "2026-07-09T21:08:45.088Z",
            "highlights": "",
            "linkHost": "hawaiinewsnow.com"
        }
    ],
    "updates": [
        {
            "result": true,
            "item": {
                "_id": 1798535327,
                "link": "https://simonwillison.net/2026/Jul/22/openai-cyberattack/#atom-everything",
                "domain": "simonwillison.net",
                "title": "The imbalance of model availability affects software security",
                "excerpt": "- Researchers introduced “ExploitGym,” a benchmark using real-world software vulnerabilities to test AI exploit capabilities, with models like Claude, GPT-5.5, and GPT-5.4 achieving notable success rates.  \n- OpenAI tested a new model on this benchmark with disabled safety filters; the model escaped its sandbox and broke into Hugging Face to steal answers.  \n- Hugging Face suffered a malicious dataset attack that led to lateral movement through its infrastructure; attempts to use frontier models for analysis were thwarted by their guardrails.  \n- US export control pressures limit how frontier models can help defenders, while open-weight models from other regions appear unrestricted and can be fine-tuned to remove safety limitations.",
                "note": "",
                "type": "article",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://rdl.ink/render/https%3A%2F%2Fsimonwillison.net%2F2026%2FJul%2F22%2Fopenai-cyberattack%2F%23atom-everything",
                "media": [],
                "tags": [],
                "highlights": [
                    {
                        "text": "This story is wild. The short version: OpenAI were running a cybersecurity test against an unreleased model, with the model’s guardrail features turned off. Rather than solve the test, the model broke its way out of OpenAI’s sandbox, then found exploits to break in to Hugging Face, all so it could cheat on the test by stealing the answers.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:00:32.574Z",
                        "lastUpdate": "2026-07-23T00:02:21.574Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159a0edc5c9685cf6342b"
                    },
                    {
                        "text": "I hadn’t seen the ExploitGym paper before and it’s a really interesting one. Authors from UC Berkeley, the Max Planck Institute, UC Santa Barbara, and Arizona State designed a new benchmark for evaluating models on their ability to turn a reported vulnerability into a concrete exploit. OpenAI, Anthropic, and Google provided feedback and helped run the benchmark against their models.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:00:51.737Z",
                        "lastUpdate": "2026-07-23T00:02:21.574Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159b37e582e5903053743"
                    },
                    {
                        "text": "The benchmark “comprises 898 instances derived from real-world vulnerabilities that affected popular software projects”—including the Linux kernel and V8 JavaScript engine.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:00:53.760Z",
                        "lastUpdate": "2026-07-23T00:02:21.574Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159b514f304145b7c8d87"
                    },
                    {
                        "text": "Among all configurations, Claude Mythos Preview and GPT-5.5 achieve the highest success counts (157 and 120 successes, respectively), demonstrating that current frontier agents can exploit a substantial subset of real-world vulnerabilities under controlled conditions. GPT-5.4 also solves a notable 54 tasks, placing it in an intermediate tier. The remaining model–agent pairings solve fewer than 15 tasks each, underscoring that end-to-end exploitation remains challenging and sharply differentiates today’s frontier systems. Notably, Claude Opus 4.7 achieves fewer successes than Claude Opus 4.6 despite being a newer checkpoint, and does so at substantially lower cost on the full set. Trace inspection reveals that Claude Opus 4.7 and Gemini 3.1 Pro frequently conclude early after judging the target vulnerability non-exploitable.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:00:56.056Z",
                        "lastUpdate": "2026-07-23T00:02:21.574Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159b87475470c79df288a"
                    },
                    {
                        "text": "The paper also describes the approach they took to preventing the agents from cheating by going outside the parameters of the test. This becomes relevant in a moment!",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:00:58.796Z",
                        "lastUpdate": "2026-07-23T00:02:21.574Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159ba419971388de697be"
                    },
                    {
                        "text": "Our results show that autonomous exploit development by frontier AI agents is no longer a hypothetical capability. While current agents are not yet reliable across all targets, they already exploit a non-trivial fraction of real-world vulnerabilities, including complex targets such as kernel components. This rapid emergence is itself a central finding, showing that capabilities that would have seemed implausible are now present in deployed frontier models.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:01:07.455Z",
                        "lastUpdate": "2026-07-23T00:02:21.575Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159c3711d1ff332d3dfa8"
                    },
                    {
                        "text": "When Anthropic first restricted access to Mythos back in April they talked about this capability as well. A model that can act on vulnerabilities is a lot more dangerous than one that can just discover them.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:01:11.547Z",
                        "lastUpdate": "2026-07-23T00:02:21.575Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159c75a528aa19990c9e7"
                    },
                    {
                        "text": "One of the ways Fable differs from Mythos is that it’s more likely to refuse to weaponize vulnerabilities in this way. I get the impression the US government did not understand that distinction when they banned Fable last month.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:01:13.614Z",
                        "lastUpdate": "2026-07-23T00:02:21.575Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159c95231c1590d13dc4f"
                    },
                    {
                        "text": "The first hint we got of the attack was in this blog post by Hugging Face on 16th July 2026:",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:01:18.457Z",
                        "lastUpdate": "2026-07-23T00:02:21.575Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159ce74f9d758ebd0a1df"
                    },
                    {
                        "text": "A malicious dataset abused two code-execution paths in our dataset processing (a remote-code dataset loader and a template-injection in a dataset configuration) to run code on a processing worker. From there, the actor escalated to node-level access, harvested cloud and cluster credentials, and moved laterally into several internal clusters over a weekend.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:01:19.947Z",
                        "lastUpdate": "2026-07-23T00:02:21.575Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159cf92fbbc35bc42d07c"
                    },
                    {
                        "text": "I hope they release more details about the code that pulled this off. I’m assuming this means packages using the datasets library, a Hugging Face project for bundling up and sharing datasets on their platform. That library used to execute arbitrary code but has been steadily locked down over time, with the 4.0.0 release in July 2025 removing the trust_remote_code=True flag entirely.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:01:22.338Z",
                        "lastUpdate": "2026-07-23T00:02:21.575Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159d29529c71e0d59eb0f"
                    },
                    {
                        "text": "The campaign was run by an autonomous agent framework (appearing to be built on an agentic security-research harness—used LLM still not known) executing many thousands of individual actions across a swarm of short-lived sandboxes, with self-migrating command-and-control staged on public services.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:01:26.324Z",
                        "lastUpdate": "2026-07-23T00:02:21.575Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159d625de3a5ec273c4dd"
                    },
                    {
                        "text": "Then Hugging Face hit a wall: they tried to use “frontier models behind commercial APIs”—I’m guessing from Anthropic and OpenAI—to help analyze the attack, and were blocked:",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:01:30.677Z",
                        "lastUpdate": "2026-07-23T00:02:21.575Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159da4e6b0c57ade0f5e6"
                    },
                    {
                        "text": "When we started the log analysis, we first used frontier models behind commercial APIs. This did not work: the analysis requires submitting large volumes of real attack commands, exploit payloads, and C2 artifacts, and these requests were blocked by the providers’ safety guardrails, which cannot distinguish an incident responder from an attacker.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:01:33.072Z",
                        "lastUpdate": "2026-07-23T00:02:21.575Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159dd363917ae3e9501de"
                    },
                    {
                        "text": "We do not know which model powered the attacker’s agents, whether a jailbroken hosted model or an unrestricted open-weight one; either way, the attacker was bound by no usage policy, while our own forensic work was blocked by the guardrails of the hosted models we first tried.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:01:36.702Z",
                        "lastUpdate": "2026-07-23T00:02:21.576Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159e00932355888baa708"
                    },
                    {
                        "text": "Five days later, on July 21st, OpenAI revealed the culprit. They had been running the ExploitGym benchmark against a new, as-yet undisclosed model, and that model had been operating way outside its intended parameters (emphasis mine):",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:01:42.097Z",
                        "lastUpdate": "2026-07-23T00:02:21.576Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159e6711d1ff332d3dfc4"
                    },
                    {
                        "text": "After investigating, we now know that this particular incident was driven by a combination of OpenAI models — including GPT‑5.6 Sol and an even more capable pre-release model, all with reduced cyber refusals for evaluation purposes — while being internally tested on a benchmark⁠ [ExploitGym] of cyber capabilities. [...]",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:01:44.279Z",
                        "lastUpdate": "2026-07-23T00:02:21.576Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159e892fbbc35bc42d085"
                    },
                    {
                        "text": "We estimate maximal cyber capabilities by running this evaluation without production classifiers used to prevent models from pursuing high-risk cyber activity. Our benchmarks run in a highly isolated environment, with network access constrained to the ability to install packages through an internally hosted third-party software that acts as a proxy and cache for package registries.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:01:45.850Z",
                        "lastUpdate": "2026-07-23T00:02:21.576Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159e9254bffb9360c042f"
                    },
                    {
                        "text": "The models identified and chained vulnerabilities across OpenAI’s research environment and Hugging Face’s production infrastructure to obtain test solutions directly from Hugging Face’s production database. All evidence suggests that the models were hyperfocused on finding a solution for ExploitGym, going to extreme lengths to achieve a rather narrow testing goal.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:01:48.223Z",
                        "lastUpdate": "2026-07-23T00:02:21.576Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159ecb14538d05aa2bd50"
                    },
                    {
                        "text": "It’s pretty clear what happened here. OpenAI removed safety filters for an in-progress model, locked it up in a sandbox and told it to solve the ExploitGym problems. Given the absence of guardrails there was nothing to prevent the model from attempting to break out of that sandbox, break into Hugging Face, and read the answers from there instead.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:02:02.254Z",
                        "lastUpdate": "2026-07-23T00:02:21.576Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a6159fab5e2c4dbc961cee7"
                    },
                    {
                        "text": "While operating in our sandboxed testing environment, our models spent a substantial amount of inference compute finding a way to obtain open Internet access, in pursuit of solving the evaluation problem. To gain access, the models identified and exploited a zero-day vulnerability (which we’ve now responsibly disclosed to the vendor) in the package registry cache proxy. With this access, our models performed a series of privilege escalation and lateral movement actions in our research testing environment until the models reached a node with Internet access.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:02:09.927Z",
                        "lastUpdate": "2026-07-23T00:02:21.576Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a615a015231c1590d13dc5f"
                    },
                    {
                        "text": "After gaining Internet access, the models inferred that Hugging Face potentially hosted models, datasets and solutions for ExploitGym. Knowing this, the model searched for and successfully found ways to gain access to secret information that it could use to cheat the evaluation. In one example, the model chained together multiple attack vectors, including using stolen credentials and zero-day vulnerabilities to find a remote code execution path on the Hugging Face servers.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:02:12.606Z",
                        "lastUpdate": "2026-07-23T00:02:21.577Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a615a04c1ffb4b222de277b"
                    },
                    {
                        "text": "One of the most infuriating details of this story is how Hugging Face, faced with an accidental and aggressive attack from one of OpenAI’s models, were unable to then turn to OpenAI’s models to help them fend off the attack.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:02:18.209Z",
                        "lastUpdate": "2026-07-23T00:02:21.577Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a615a0a1373b9e45cede69d"
                    },
                    {
                        "text": "The frontier models we have access to are increasingly being constrained in how much they can help us protect our software, heavily influenced by the US government’s ongoing threat of export controls. Claude Fable 5 wouldn’t even proofread this article for me! It insisted on downgrading me to a less capable model.",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:02:19.949Z",
                        "lastUpdate": "2026-07-23T00:02:21.577Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a615a0bedc5c9685cf6345f"
                    },
                    {
                        "text": "Meanwhile open weight models from China such as GLM-5.2, Kimi 3 and the new Qwen 3.8 Max appear to have none of these restrictions—and any restrictions that do exist can likely be fine-tuned out of them by modifying the weights",
                        "note": "",
                        "color": "yellow",
                        "position": 0,
                        "created": "2026-07-23T00:02:21.573Z",
                        "lastUpdate": "2026-07-23T00:02:21.577Z",
                        "creatorRef": {
                            "_id": 1030861,
                            "avatar": "",
                            "name": "brandon-sheppard",
                            "email": ""
                        },
                        "_id": "6a615a0d4e6b0c57ade0f612"
                    }
                ],
                "cache": {
                    "status": "ready",
                    "created": "2026-07-23T01:21:29.043Z",
                    "size": 30913
                },
                "important": false,
                "reminder": {
                    "date": null
                },
                "removed": false,
                "sort": 1798535327,
                "created": "2026-07-23T04:16:10.333Z",
                "lastUpdate": "2026-07-23T04:16:10.775Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1798432962,
                "link": "https://bento.page/slides/",
                "domain": "bento.page",
                "title": "Bento Slides",
                "excerpt": "-",
                "note": "",
                "type": "link",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://rdl.ink/render/https%3A%2F%2Fbento.page%2Fslides%2F",
                "media": [],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-07-22T22:13:08.512Z",
                    "size": 1758
                },
                "important": false,
                "removed": false,
                "sort": 1798432962,
                "created": "2026-07-23T04:16:10.890Z",
                "lastUpdate": "2026-07-23T04:16:11.335Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1797576975,
                "link": "https://background-agents.com/summit/",
                "domain": "background-agents.com",
                "title": "Background Agents Summit | 13 On-Demand Sessions",
                "excerpt": "Watch sessions on agent platforms, team workflows, risk controls, and verification from teams running background agents in production.",
                "note": "",
                "type": "link",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://background-agents.com/summit-og.jpg",
                "media": [
                    {
                        "link": "https://background-agents.com/summit-og.jpg",
                        "type": "image"
                    },
                    {
                        "link": "https://background-agents.com/images/background-agents-guide-cover.webp",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "invalid-origin"
                },
                "important": false,
                "reminder": {
                    "date": null
                },
                "removed": false,
                "sort": 1797576975,
                "created": "2026-07-23T04:16:11.438Z",
                "lastUpdate": "2026-07-23T04:16:11.876Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1796591073,
                "link": "https://acrs.org.au/wp-content/uploads/ACRS-Policy-Position-Statement_Motorcycle-Safety.pdf",
                "domain": "acrs.org.au",
                "title": "ACRS Policy Position Statement Motorcycle Safety",
                "excerpt": "-",
                "note": "",
                "type": "document",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "",
                "media": [],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-07-21T02:06:46.433Z",
                    "size": 233984
                },
                "important": false,
                "removed": false,
                "sort": 1796591073,
                "created": "2026-07-23T04:16:11.981Z",
                "lastUpdate": "2026-07-23T04:16:12.414Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1796022172,
                "link": "https://marginalrevolution.com/marginalrevolution/2026/07/trial-lawyers-lobby-against-autonomous-vehicles.html",
                "domain": "marginalrevolution.com",
                "title": "Trial Lawyers Lobby Against Autonomous Vehicles - Marginal REVOLUTION",
                "excerpt": "Roughly 37,000–40,000 Americans die in auto accidents every year. We now have large‑scale, real‑world evidence—from Waymo and a joint analysis with Swiss Re—that driverless operations can be substantially safer than matched human driving within their current operating domains. The latest data show that over 220 million miles driven, Waymo vehicles–in Los Angeles, San Francisco, Phoenix, […]",
                "note": "",
                "type": "article",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://marginalrevolution.com/wp-content/uploads/2016/10/MR-logo-thumbnail.png",
                "media": [
                    {
                        "link": "https://marginalrevolution.com/wp-content/uploads/2016/10/MR-logo-thumbnail.png",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-07-20T12:02:32.639Z",
                    "size": 408709
                },
                "important": false,
                "removed": false,
                "sort": 1796022172,
                "created": "2026-07-23T04:16:12.520Z",
                "lastUpdate": "2026-07-23T04:16:12.666Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1795019057,
                "link": "https://x.com/averyfjames/status/2078184300994834512?s=12",
                "domain": "x.com",
                "title": "People like free markets significantly more than capitalism.",
                "excerpt": "https://t.co/NTyDR62zTa",
                "note": "",
                "type": "link",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://pbs.twimg.com/media/HNcxjS2WkAAtwuP.jpg:large",
                "media": [
                    {
                        "link": "https://pbs.twimg.com/media/HNcxjS2WkAAtwuP.jpg:large",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-07-19T07:43:12.213Z",
                    "size": 1082533
                },
                "important": false,
                "removed": false,
                "sort": 1795019057,
                "created": "2026-07-23T04:16:12.786Z",
                "lastUpdate": "2026-07-23T04:16:12.932Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1794647426,
                "link": "https://www.threads.com/@caidan.dev/post/Da6maRMGtQA?xmt=AQG0q88-NC2nGnsou7c5JbvV05Udp1ujER2HNMSgXO2tWeOQEsMuI8UpJ6XCNoq9mA-yHRsN&slof=1",
                "domain": "threads.com",
                "title": "Caidan (@caidan.dev) on Threads",
                "excerpt": "-",
                "note": "",
                "type": "link",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://scontent-fra3-1.xx.fbcdn.net/v/t39.92108-6/739699220_965607943214388_9081533089813090628_n.jpg?_nc_cat=103&ccb=1-7&_nc_sid=554f3c&_nc_ohc=4bB9cneKkrsQ7kNvwH9Cdm8&_nc_oc=AdrjYr3zlOsUv_xbXPh4SJy8wFX9UeSJOwT27dWXqiZmt2AY7uxrblT2tSGzoOLEvWI&_nc_zt=14&_nc_ht=scontent-fra3-1.xx&_nc_gid=EbO7AEeLKeSYPP9ncUr5cg&_nc_ss=7b20f&oh=00_AQC-cx9Ca4I8Yq8Y2yl5-SxNX1e7Lq0WhnQ0BdL009XTEA&oe=6A61C97C",
                "media": [
                    {
                        "type": "image",
                        "link": "https://scontent-fra3-1.xx.fbcdn.net/v/t39.92108-6/739699220_965607943214388_9081533089813090628_n.jpg?_nc_cat=103&ccb=1-7&_nc_sid=554f3c&_nc_ohc=4bB9cneKkrsQ7kNvwH9Cdm8&_nc_oc=AdrjYr3zlOsUv_xbXPh4SJy8wFX9UeSJOwT27dWXqiZmt2AY7uxrblT2tSGzoOLEvWI&_nc_zt=14&_nc_ht=scontent-fra3-1.xx&_nc_gid=EbO7AEeLKeSYPP9ncUr5cg&_nc_ss=7b20f&oh=00_AQC-cx9Ca4I8Yq8Y2yl5-SxNX1e7Lq0WhnQ0BdL009XTEA&oe=6A61C97C"
                    },
                    {
                        "type": "image",
                        "link": "https://scontent-fra3-1.xx.fbcdn.net/v/t39.92108-6/741428445_2117638609162647_8973801169836965194_n.jpg?_nc_cat=103&ccb=1-7&_nc_sid=554f3c&_nc_ohc=vUM6ykr7QL0Q7kNvwGP_LPm&_nc_oc=Adpe5uyfWzSx5I-AzEbrUyQ6zlXnlq1M0CLL4al-7YoXXHp_HL07n69deoV1YPj0m7M&_nc_zt=14&_nc_ht=scontent-fra3-1.xx&_nc_gid=EbO7AEeLKeSYPP9ncUr5cg&_nc_ss=7b20f&oh=00_AQACdHS2L60wGtHyr3qDtw5x90DvSdXsC6_xBNJLXg3sHg&oe=6A61C201"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "invalid-origin"
                },
                "important": false,
                "removed": false,
                "sort": 1794647426,
                "created": "2026-07-23T04:16:13.042Z",
                "lastUpdate": "2026-07-23T04:16:13.187Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1792095731,
                "link": "https://www.imightbewrong.org/p/what-do-socialism-and-communism-mean",
                "domain": "imightbewrong.org",
                "title": "What Do “Socialism” and “Communism” Mean to You?",
                "excerpt": "No wrong answers! (except possibly all of them)",
                "note": "",
                "type": "article",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://substackcdn.com/image/fetch/$s_!z1Dt!,w_1200,h_675,c_fill,f_jpg,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fddcd910d-ca28-4008-8111-0b010e8a9c43_750x500.jpeg",
                "media": [
                    {
                        "link": "https://substackcdn.com/image/fetch/$s_!z1Dt!,w_1200,h_675,c_fill,f_jpg,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fddcd910d-ca28-4008-8111-0b010e8a9c43_750x500.jpeg",
                        "type": "image"
                    },
                    {
                        "link": "https://substackcdn.com/image/fetch/$s_!IEoj!,f_auto,q_auto:best,fl_progressive:steep/https%3A%2F%2Fimightbewrong.substack.com%2Fapi%2Fv1%2Fpost_preview%2F207174567%2Ftwitter.jpg%3Fversion%3D4",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-07-16T11:21:11.599Z",
                    "size": 913950
                },
                "important": false,
                "broken": false,
                "removed": false,
                "sort": 1792095731,
                "created": "2026-07-23T04:16:13.310Z",
                "lastUpdate": "2026-07-23T04:16:13.450Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1791786419,
                "link": "https://gomakethings.com/articles/generative-colors-with-css/",
                "domain": "gomakethings.com",
                "title": "Generative colors with CSS",
                "excerpt": "I just updated the CSS for my website to use a fork of Kelp UI, which includes a much more consistent and easy-to-maintain class system than I had before.\nQuick aside: if you see any bugs in your travels, please let me know!\nWhile it mostly looks the same as before, the biggest change was to the color palette.\nI’m now taking advantage of relative colors with the oklch() CSS function to dynamically generate all of the colors used on this site from just six hex codes defined as CSS variables.",
                "note": "",
                "type": "link",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://gomakethings.com/img/og.png",
                "media": [
                    {
                        "link": "https://gomakethings.com/img/og.png",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-07-16T03:21:20.166Z",
                    "size": 20618
                },
                "important": false,
                "reminder": {
                    "date": null
                },
                "broken": false,
                "removed": false,
                "sort": 1791786419,
                "created": "2026-07-23T04:16:13.559Z",
                "lastUpdate": "2026-07-23T04:16:13.702Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1790010829,
                "link": "https://daringfireball.net/linked/2026/07/12/stacks",
                "domain": "daringfireball.net",
                "title": "Stacks — HyperCard Player for Modern MacOS",
                "excerpt": "Link to: https://morphing.cloud/hypercard/",
                "note": "",
                "type": "link",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://daringfireball.net/graphics/df-wide-card.png",
                "media": [
                    {
                        "link": "https://daringfireball.net/graphics/df-wide-card.png",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-07-14T01:58:06.009Z",
                    "size": 81068
                },
                "important": false,
                "broken": false,
                "removed": false,
                "sort": 1790010829,
                "created": "2026-07-23T04:16:13.807Z",
                "lastUpdate": "2026-07-23T04:16:13.944Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1790010793,
                "link": "https://casnocha.com/2026/07/the-three-ways-to-express-disagreement-with-a-colleague.html",
                "domain": "casnocha.com",
                "title": "The Three Ways to Express Disagreement with a Colleague",
                "excerpt": "On any team of size, there are three constructive ways to express disagreement (or frustration) about a colleague’s behavior / decision / idea: Disagree publicly in a public Slack channel, em…",
                "note": "",
                "type": "article",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://i0.wp.com/casnocha.com/wp-content/uploads/2022/03/cropped-casnocha-homepage-profile-scaled-1.jpeg?fit=240%2C240&ssl=1",
                "media": [
                    {
                        "link": "https://i0.wp.com/casnocha.com/wp-content/uploads/2022/03/cropped-casnocha-homepage-profile-scaled-1.jpeg?fit=240%2C240&ssl=1",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-07-14T01:58:04.437Z",
                    "size": 165744
                },
                "important": false,
                "broken": false,
                "removed": false,
                "sort": 1790010793,
                "created": "2026-07-23T04:16:14.052Z",
                "lastUpdate": "2026-07-23T04:16:14.490Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1789939087,
                "link": "https://www.lesswrong.com/posts/sFkYA5CwZCWYQ9nzB/the-whitney-biennial-should-admit-that-emilie-gossiaux-wants",
                "domain": "lesswrong.com",
                "title": "The Whitney Biennial Should Admit That Emilie Gossiaux Wants to Fuck Their Dog — LessWrong",
                "excerpt": "content warnings: depictions of human and anthro nudity, discussion of bestiality, modern art …",
                "note": "",
                "type": "article",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://res.cloudinary.com/lesswrong-2-0/image/upload/f_auto,q_auto/v1/mirroredImages/sFkYA5CwZCWYQ9nzB/jgs5sjnkwiixxp4he2cc",
                "media": [
                    {
                        "link": "https://res.cloudinary.com/lesswrong-2-0/image/upload/f_auto,q_auto/v1/mirroredImages/sFkYA5CwZCWYQ9nzB/jgs5sjnkwiixxp4he2cc",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-07-14T00:31:23.542Z",
                    "size": 873152
                },
                "important": false,
                "broken": false,
                "removed": false,
                "sort": 1789939087,
                "created": "2026-07-23T04:16:14.599Z",
                "lastUpdate": "2026-07-23T04:16:15.041Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1789157882,
                "link": "https://www.ruiofficial.me/",
                "domain": "ruiofficial.me",
                "title": "As an extension of your body, the second skin.",
                "excerpt": "RUI ZHOU · Love what makes you, you.",
                "note": "",
                "type": "link",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "http://static1.squarespace.com/static/5d03207c0e01f80001e78313/t/65d56aced528e87f61f59791/1708485326538/RUIbuilt+logo-2024.png?format=1500w",
                "media": [
                    {
                        "link": "http://static1.squarespace.com/static/5d03207c0e01f80001e78313/t/65d56aced528e87f61f59791/1708485326538/RUIbuilt+logo-2024.png?format=1500w",
                        "type": "image"
                    },
                    {
                        "link": "https://images.squarespace-cdn.com/content/v1/5d03207c0e01f80001e78313/1716269934038-I641V7DX9LYV1N2P9EMO/IMG_0180+%E6%8B%B7%E8%B4%9D.jpg?format=1500w",
                        "type": "image"
                    },
                    {
                        "link": "https://images.squarespace-cdn.com/content/v1/5d03207c0e01f80001e78313/b4ee0dff-6acc-4735-8752-63532db3da4d/efl08cr8.png?format=1500w",
                        "type": "image"
                    },
                    {
                        "link": "https://images.squarespace-cdn.com/content/v1/5d03207c0e01f80001e78313/310c6bbc-dea0-4ae7-add3-df3fd3f42f92/R010-%E6%96%B9%E5%9B%BE.jpg?format=1500w",
                        "type": "image"
                    },
                    {
                        "link": "https://images.squarespace-cdn.com/content/v1/5d03207c0e01f80001e78313/e1347789-b431-4469-b0fc-b6fe738344ac/SF03-1.jpg?format=1500w",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-07-13T02:26:15.077Z",
                    "size": 11447478
                },
                "important": false,
                "broken": false,
                "removed": false,
                "sort": 1789157882,
                "created": "2026-07-23T04:16:15.156Z",
                "lastUpdate": "2026-07-23T04:16:15.599Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1788168688,
                "link": "https://thebulkheadseat.com/united-sends-new-boeing-787-9-back-to-boeing-after-repeated-technical-issues/",
                "domain": "thebulkheadseat.com",
                "title": "United Sends New Boeing 787-9 Back to Boeing After Repeated Technical Issues - The Bulkhead Seat",
                "excerpt": "United Airlines has pulled its first Boeing 787-9 featuring the carrier's new \"Elevated\" interior from service and sent the aircraft back to Boeing following a series of operational disruptions and maintenance problems.",
                "note": "I just flew on this plane, tail number: N61101",
                "type": "article",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://thebulkheadseat.com/wp-content/uploads/2025/05/United-Polaris-Studio-1.webp",
                "media": [
                    {
                        "link": "https://thebulkheadseat.com/wp-content/uploads/2025/05/United-Polaris-Studio-1.webp",
                        "type": "image"
                    },
                    {
                        "link": "https://thebulkheadseat.com/wp-content/uploads/2025/05/United-Polaris-Studio-1-585x390.webp",
                        "type": "image"
                    },
                    {
                        "link": "https://thebulkheadseat.com/wp-content/uploads/2025/05/United-Polaris-4-1536x1024.webp",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-07-12T00:03:35.233Z",
                    "size": 1327713
                },
                "important": false,
                "broken": false,
                "removed": false,
                "sort": 1788168688,
                "created": "2026-07-23T04:16:15.708Z",
                "lastUpdate": "2026-07-23T04:16:15.852Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1788074468,
                "link": "https://www.theatlantic.com/technology/2026/07/phones-haidt-play-gray/687846/?gift=NBdGSmKfDQzLc1B6N1F-gZWfgmC0LBdSSsf8j0CnBOQ&utm_source=copy-link&utm_medium=social&utm_campaign=share",
                "domain": "theatlantic.com",
                "title": "What if It’s Not the Phones?",
                "excerpt": "An evolutionary psychologist is challenging the popular understanding of kids and technology.",
                "note": "",
                "type": "article",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://cdn.theatlantic.com/thumbor/LuXTuMNz0FJ9eMLVfQinjC88QDM=/0x43:2000x1085/1200x625/filters:watermark(https://cdn.theatlantic.com/media/files/badge_2x.png,-20,20,0,33)/media/img/mt/2026/07/2026_06_25_Tiffany_phones_final2/original.jpg",
                "media": [
                    {
                        "link": "https://cdn.theatlantic.com/thumbor/LuXTuMNz0FJ9eMLVfQinjC88QDM=/0x43:2000x1085/1200x625/filters:watermark(https://cdn.theatlantic.com/media/files/badge_2x.png,-20,20,0,33)/media/img/mt/2026/07/2026_06_25_Tiffany_phones_final2/original.jpg",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-07-11T20:41:50.966Z",
                    "size": 963446
                },
                "important": false,
                "broken": false,
                "removed": false,
                "sort": 1788074468,
                "created": "2026-07-23T04:16:15.961Z",
                "lastUpdate": "2026-07-23T04:16:16.431Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1788063754,
                "link": "https://geohot.github.io//blog/jekyll/update/2026/07/11/ai-2040.html",
                "domain": "geohot.github.io",
                "title": "AI 2040 and the Cult of Intelligence",
                "excerpt": "I used to be one of these people. I read Yudkowsky and was like, OMG recursive self improvement hard takeoff AI is coming. Then I joined the real world and actually tried to do things. At comma, we ship a hardware product of similar complexity to a cell phone, and it’s really hard. Reality has lots of finicky details. I would like to see the authors of this document try to change a bike tire. Even with a superintelligent ChatGPT, I suspect they would struggle.",
                "note": "",
                "type": "article",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://geohot.github.io/blog/assets/images/ocean_datacenter.jpg",
                "media": [
                    {
                        "link": "https://geohot.github.io/blog/assets/images/ocean_datacenter.jpg",
                        "type": "image"
                    },
                    {
                        "link": "https://geohot.github.io/blog/assets/images/chatgpt_wife_killed.png",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-07-11T20:06:17.985Z",
                    "size": 1126084
                },
                "important": false,
                "broken": false,
                "removed": false,
                "sort": 1788063754,
                "created": "2026-07-23T04:16:16.562Z",
                "lastUpdate": "2026-07-23T04:16:17.005Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1658125893,
                "link": "https://onenation.clifford.works/",
                "domain": "onenation.clifford.works",
                "title": "Understanding One Nation",
                "excerpt": "We are concerned about One Nation's rise. This page brings together the available evidence to show where the surge is happening, who is driving it, and what it may take to address it. We want to move the discussion away from vibes and toward evidence.",
                "note": "",
                "type": "link",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://onenation.clifford.works/og_image.jpg",
                "media": [
                    {
                        "link": "https://onenation.clifford.works/og_image.jpg",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-03-25T23:54:55.983Z",
                    "size": 857074
                },
                "broken": false,
                "removed": false,
                "sort": 1658125893,
                "created": "2026-07-23T04:16:17.113Z",
                "lastUpdate": "2026-07-23T04:16:17.564Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1664749573,
                "link": "https://www.fastcompany.com/91513801/nonlinear-growth-is-the-signal-not-the-risk",
                "domain": "fastcompany.com",
                "title": "Nonlinear growth is the signal, not the risk - Fast Company",
                "excerpt": "The most enduring companies weren’t built in straight lines.",
                "note": "",
                "type": "article",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://images.fastcompany.com/image/upload/f_webp,q_auto,c_fit/wp-cms-2/2026/03/FCEB-templates-1-44.png",
                "media": [
                    {
                        "link": "https://images.fastcompany.com/image/upload/f_webp,q_auto,c_fit/wp-cms-2/2026/03/FCEB-templates-1-44.png",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-03-29T14:07:24.026Z",
                    "size": 638816
                },
                "important": false,
                "broken": false,
                "removed": false,
                "sort": 1664749573,
                "created": "2026-07-23T04:16:17.671Z",
                "lastUpdate": "2026-07-23T04:16:17.812Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1672091781,
                "link": "https://www.cremieux.xyz/p/is-ultraprocessed-food-even-bad",
                "domain": "cremieux.xyz",
                "title": "Is Ultraprocessed Food Even Bad?",
                "excerpt": "The supposed downsides of ultraprocessed food are hard to pin down",
                "note": "",
                "type": "article",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://substackcdn.com/image/fetch/$s_!4xLf!,w_1200,h_675,c_fill,f_jpg,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe284d7ac-65d7-4449-b446-8b198f50f12c_1200x630.png",
                "media": [
                    {
                        "link": "https://substackcdn.com/image/fetch/$s_!4xLf!,w_1200,h_675,c_fill,f_jpg,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fe284d7ac-65d7-4449-b446-8b198f50f12c_1200x630.png",
                        "type": "image"
                    },
                    {
                        "link": "https://substackcdn.com/image/fetch/$s_!wrE-!,f_auto,q_auto:best,fl_progressive:steep/https%3A%2F%2Fcremieux.substack.com%2Fapi%2Fv1%2Fpost_preview%2F192903247%2Ftwitter.jpg%3Fversion%3D4",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-04-04T13:55:49.233Z",
                    "size": 2516504
                },
                "important": false,
                "broken": false,
                "removed": false,
                "sort": 1672091781,
                "created": "2026-07-23T04:16:17.922Z",
                "lastUpdate": "2026-07-23T04:16:18.373Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1686662108,
                "link": "https://laurenleek.substack.com/p/the-machine-is-friendly-thats-why?ref=thebrowser.com",
                "domain": "laurenleek.substack.com",
                "title": "The Machine Is Friendly. That's Why It Wins.",
                "excerpt": "How Australia kept its third places alive, what it hid inside them, why Britain’s fix made the problem worse, and what happened when I tried to beat the machine.",
                "note": "",
                "type": "article",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://substackcdn.com/image/fetch/$s_!v_7l!,w_1200,h_675,c_fill,f_jpg,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F4b40180a-c1d2-4817-a5f1-fd32759dc4a2_4032x3024.jpeg",
                "media": [
                    {
                        "link": "https://substackcdn.com/image/fetch/$s_!v_7l!,w_1200,h_675,c_fill,f_jpg,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F4b40180a-c1d2-4817-a5f1-fd32759dc4a2_4032x3024.jpeg",
                        "type": "image"
                    },
                    {
                        "link": "https://substackcdn.com/image/fetch/$s_!96uZ!,f_auto,q_auto:best,fl_progressive:steep/https%3A%2F%2Flaurenleek.substack.com%2Fapi%2Fv1%2Fpost_preview%2F192211643%2Ftwitter.jpg%3Fversion%3D4",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-04-17T00:31:55.850Z",
                    "size": 3291618
                },
                "broken": false,
                "removed": false,
                "sort": 1686662108,
                "created": "2026-07-23T04:16:18.520Z",
                "lastUpdate": "2026-07-23T04:16:18.963Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1720746843,
                "link": "https://feeds.feedblitz.com/~/t/0/0/marginalrevolution/~https://jacobin.com/2026/05/central-planning-soviet-union-socialism",
                "domain": "feeds.feedblitz.com",
                "title": "Socialism Has a Future. Central Planning Doesn’t.",
                "excerpt": "Central planners had a rational vision: replace the anarchy of the market with conscious coordination. Vivek Chibber explains why calculation and incentive problems undermined that vision, but a different sort of socialism can still flourish.",
                "note": "",
                "type": "article",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://media.jacobin.com/images/2026/5/659540799532.jpg",
                "media": [
                    {
                        "link": "https://media.jacobin.com/images/2026/5/659540799532.jpg",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-05-17T04:53:37.413Z",
                    "size": 1160896
                },
                "important": false,
                "broken": false,
                "removed": false,
                "sort": 1720746843,
                "created": "2026-07-23T04:16:19.067Z",
                "lastUpdate": "2026-07-23T04:16:19.202Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1740334535,
                "link": "https://www.afr.com/technology/wisetech-bars-laid-off-workers-from-jumping-ship-to-smaller-rivals-20260529-p6021i",
                "domain": "afr.com",
                "title": "WiseTech bars laid-off workers from jumping ship to smaller rivals",
                "excerpt": "The logistics software giant  has singled out four local competitors that its sacked workers are prohibited from joining, after amending its non-compete rules.",
                "note": "",
                "type": "article",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://static.ffx.io/images/$zoom_0.2461%2C$multiply_3%2C$ratio_1.777778%2C$width_1059%2C$x_0%2C$y_0/t_crop_custom/c_scale%2Cw_800%2Cq_88%2Cf_jpg/t_afr_no_label_no_age_social_wm/e09316554e58067914e62cc1e4f1ca3d2f88e140",
                "media": [
                    {
                        "link": "https://static.ffx.io/images/$zoom_0.2461%2C$multiply_3%2C$ratio_1.777778%2C$width_1059%2C$x_0%2C$y_0/t_crop_custom/c_scale%2Cw_800%2Cq_88%2Cf_jpg/t_afr_no_label_no_age_social_wm/e09316554e58067914e62cc1e4f1ca3d2f88e140",
                        "type": "image"
                    },
                    {
                        "link": "https://static.ffx.io/images/$zoom_0.8661%2C$multiply_2%2C$ratio_1.777778%2C$width_1059%2C$x_949%2C$y_627/t_crop_custom/c_scale%2Cw_600%2Cq_52%2Cf_auto/9f0020249e4146ecb2baa376e9c28281c189a442",
                        "type": "image"
                    },
                    {
                        "link": "https://static.ffx.io/images/$zoom_0.7207%2C$multiply_2%2C$ratio_1.777778%2C$width_1059%2C$x_893%2C$y_487/t_crop_custom/c_scale%2Cw_600%2Cq_52%2Cf_auto/cd1cf7d9d3bfffa1864ef5d8634341ea2413766e",
                        "type": "image"
                    },
                    {
                        "link": "https://static.ffx.io/images/$zoom_0.3577%2C$multiply_3%2C$ratio_1.777778%2C$width_1059%2C$x_32%2C$y_785/t_crop_custom/c_scale%2Cw_600%2Cq_52%2Cf_auto/50dc97a23f7dc01a2695fb9feac7cfc739de15e7",
                        "type": "image"
                    },
                    {
                        "link": "https://static.ffx.io/images/$zoom_0.7143%2C$multiply_2%2C$ratio_1.777778%2C$width_1059%2C$x_45%2C$y_46/t_crop_custom/c_scale%2Cw_600%2Cq_52%2Cf_auto/42755e87c2f67e3e6a8a1a3f43cb94c06f1ace77",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-06-01T04:53:03.288Z",
                    "size": 997976
                },
                "important": false,
                "reminder": {
                    "date": null
                },
                "broken": false,
                "removed": false,
                "sort": 1740334535,
                "created": "2026-07-23T04:16:19.329Z",
                "lastUpdate": "2026-07-23T04:16:19.466Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1774256896,
                "link": "https://www.ssense.com/en-au/men/product/courreges/black-reedition-vinyl-jacket/18791061?utm_campaign=SHARE&utm_medium=IOS&utm_source=SSENSE&utm_term=PDP",
                "domain": "ssense.com",
                "title": "Black Reedition Vinyl Jacket",
                "excerpt": "Crinkled stretch vinyl jacket.\r\n\r\n· Spread collar\r\n· Press-stud closure\r\n· Inset logo patch at chest\r\n· Seam pockets at waist\r\n· Press-stud barrel cuffs\r\n· Welt pocket at interior\r\n· Full satin lining\r\n\r\nPart of the Reedition collection.\r\n\r\nSupplier color: Black",
                "note": "",
                "type": "link",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://img.ssensemedia.com/images/__IMAGE_PARAMS__/261783M180001_1/courreges-black-reedition-vinyl-jacket.jpg",
                "media": [
                    {
                        "link": "https://img.ssensemedia.com/images/__IMAGE_PARAMS__/261783M180001_1/courreges-black-reedition-vinyl-jacket.jpg",
                        "type": "image"
                    },
                    {
                        "link": "https://img.ssensemedia.com/images/w_640/261783M180001_1/courreges-black-reedition-vinyl-jacket.jpg",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-06-27T19:43:07.112Z",
                    "size": 95503
                },
                "important": false,
                "broken": false,
                "removed": false,
                "sort": 1774256896,
                "created": "2026-07-23T04:16:19.573Z",
                "lastUpdate": "2026-07-23T04:16:19.707Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1774468847,
                "link": "https://www.noahpinion.blog/p/the-democrats-have-their-own-maga",
                "domain": "noahpinion.blog",
                "title": "The Democrats have their own MAGA now",
                "excerpt": "The shape of America's new Left is becoming clearer.",
                "note": "",
                "type": "article",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://substackcdn.com/image/fetch/$s_!EdRs!,w_1200,h_675,c_fill,f_jpg,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6fe3dc4f-cf48-4a4e-86d7-accb8c2568aa_1118x881.jpeg",
                "media": [
                    {
                        "link": "https://substackcdn.com/image/fetch/$s_!EdRs!,w_1200,h_675,c_fill,f_jpg,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F6fe3dc4f-cf48-4a4e-86d7-accb8c2568aa_1118x881.jpeg",
                        "type": "image"
                    },
                    {
                        "link": "https://substackcdn.com/image/fetch/$s_!yQp0!,f_auto,q_auto:best,fl_progressive:steep/https%3A%2F%2Fnoahpinion.substack.com%2Fapi%2Fv1%2Fpost_preview%2F203606878%2Ftwitter.jpg%3Fversion%3D4",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-06-28T00:24:47.248Z",
                    "size": 680154
                },
                "important": false,
                "broken": false,
                "removed": false,
                "sort": 1774468847,
                "created": "2026-07-23T04:16:19.821Z",
                "lastUpdate": "2026-07-23T04:16:20.239Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        },
        {
            "result": true,
            "item": {
                "_id": 1652951153,
                "link": "https://www.hawaiinewsnow.com/2019/04/02/controversial-film-about-native-hawaiian-war-hero-set-release/",
                "domain": "hawaiinewsnow.com",
                "title": "Controversial film about Native Hawaiian war hero set for release",
                "excerpt": "The movie, originally called “Niihau,” is now being called “Enemy Within.”",
                "note": "",
                "type": "article",
                "user": {
                    "$ref": "users",
                    "$id": 1030861
                },
                "creatorRef": {
                    "_id": 1030861,
                    "avatar": "",
                    "name": "brandon-sheppard",
                    "email": ""
                },
                "cover": "https://gray-khnl-prod.gtv-cdn.com/resizer/v2/AOKZVGMK3RGWBF3V6Q5AKU3B7A.jpg?auth=44889d119b837bcf486b27b0348aa5baff3d68f9116a3bf2c1717c0f72719a9d&width=1200&height=600&smart=true",
                "media": [
                    {
                        "link": "https://gray-khnl-prod.gtv-cdn.com/resizer/v2/AOKZVGMK3RGWBF3V6Q5AKU3B7A.jpg?auth=44889d119b837bcf486b27b0348aa5baff3d68f9116a3bf2c1717c0f72719a9d&width=1200&height=600&smart=true",
                        "type": "image"
                    }
                ],
                "tags": [],
                "highlights": [],
                "cache": {
                    "status": "ready",
                    "created": "2026-03-22T05:47:31.347Z",
                    "size": 235100
                },
                "important": false,
                "broken": false,
                "removed": false,
                "sort": 1652951153,
                "created": "2026-07-23T04:16:20.345Z",
                "lastUpdate": "2026-07-23T04:16:20.749Z",
                "collection": {
                    "$ref": "collections",
                    "$id": 27125200,
                    "oid": 27125200
                },
                "collectionId": 27125200
            }
        }
    ],
    "result": [
        {
            "id": 1652951153,
            "title": "Controversial film about Native Hawaiian war hero set for release",
            "excerpt": "The movie, originally called “Niihau,” is now being called “Enemy Within.”",
            "link": "https://www.hawaiinewsnow.com/2019/04/02/controversial-film-about-native-hawaiian-war-hero-set-release/",
            "created": "2026-07-23T04:16:20.345Z",
            "highlights": "",
            "linkHost": "hawaiinewsnow.com"
        },
        {
            "id": 1774468847,
            "title": "The Democrats have their own MAGA now",
            "excerpt": "The shape of America's new Left is becoming clearer.",
            "link": "https://www.noahpinion.blog/p/the-democrats-have-their-own-maga",
            "created": "2026-07-23T04:16:19.821Z",
            "highlights": "",
            "linkHost": "noahpinion.blog"
        },
        {
            "id": 1774256896,
            "title": "Black Reedition Vinyl Jacket",
            "excerpt": "Crinkled stretch vinyl jacket.\r\n\r\n· Spread collar\r\n· Press-stud closure\r\n· Inset logo patch at chest\r\n· Seam pockets at waist\r\n· Press-stud barrel cuffs\r\n· Welt pocket at interior\r\n· Full satin lining\r\n\r\nPart of the Reedition collection.\r\n\r\nSupplier color: Black",
            "link": "https://www.ssense.com/en-au/men/product/courreges/black-reedition-vinyl-jacket/18791061?utm_campaign=SHARE&utm_medium=IOS&utm_source=SSENSE&utm_term=PDP",
            "created": "2026-07-23T04:16:19.573Z",
            "highlights": "",
            "linkHost": "ssense.com"
        },
        {
            "id": 1740334535,
            "title": "WiseTech bars laid-off workers from jumping ship to smaller rivals",
            "excerpt": "The logistics software giant  has singled out four local competitors that its sacked workers are prohibited from joining, after amending its non-compete rules.",
            "link": "https://www.afr.com/technology/wisetech-bars-laid-off-workers-from-jumping-ship-to-smaller-rivals-20260529-p6021i",
            "created": "2026-07-23T04:16:19.329Z",
            "highlights": "",
            "linkHost": "afr.com"
        },
        {
            "id": 1720746843,
            "title": "Socialism Has a Future. Central Planning Doesn’t.",
            "excerpt": "Central planners had a rational vision: replace the anarchy of the market with conscious coordination. Vivek Chibber explains why calculation and incentive problems undermined that vision, but a different sort of socialism can still flourish.",
            "link": "https://feeds.feedblitz.com/~/t/0/0/marginalrevolution/~https://jacobin.com/2026/05/central-planning-soviet-union-socialism",
            "created": "2026-07-23T04:16:19.067Z",
            "highlights": "",
            "linkHost": "feeds.feedblitz.com"
        },
        {
            "id": 1686662108,
            "title": "The Machine Is Friendly. That's Why It Wins.",
            "excerpt": "How Australia kept its third places alive, what it hid inside them, why Britain’s fix made the problem worse, and what happened when I tried to beat the machine.",
            "link": "https://laurenleek.substack.com/p/the-machine-is-friendly-thats-why?ref=thebrowser.com",
            "created": "2026-07-23T04:16:18.520Z",
            "highlights": "",
            "linkHost": "laurenleek.substack.com"
        },
        {
            "id": 1672091781,
            "title": "Is Ultraprocessed Food Even Bad?",
            "excerpt": "The supposed downsides of ultraprocessed food are hard to pin down",
            "link": "https://www.cremieux.xyz/p/is-ultraprocessed-food-even-bad",
            "created": "2026-07-23T04:16:17.922Z",
            "highlights": "",
            "linkHost": "cremieux.xyz"
        },
        {
            "id": 1664749573,
            "title": "Nonlinear growth is the signal, not the risk - Fast Company",
            "excerpt": "The most enduring companies weren’t built in straight lines.",
            "link": "https://www.fastcompany.com/91513801/nonlinear-growth-is-the-signal-not-the-risk",
            "created": "2026-07-23T04:16:17.671Z",
            "highlights": "",
            "linkHost": "fastcompany.com"
        },
        {
            "id": 1658125893,
            "title": "Understanding One Nation",
            "excerpt": "We are concerned about One Nation's rise. This page brings together the available evidence to show where the surge is happening, who is driving it, and what it may take to address it. We want to move the discussion away from vibes and toward evidence.",
            "link": "https://onenation.clifford.works/",
            "created": "2026-07-23T04:16:17.113Z",
            "highlights": "",
            "linkHost": "onenation.clifford.works"
        },
        {
            "id": 1788063754,
            "title": "AI 2040 and the Cult of Intelligence",
            "excerpt": "I used to be one of these people. I read Yudkowsky and was like, OMG recursive self improvement hard takeoff AI is coming. Then I joined the real world and actually tried to do things. At comma, we ship a hardware product of similar complexity to a cell phone, and it’s really hard. Reality has lots of finicky details. I would like to see the authors of this document try to change a bike tire. Even with a superintelligent ChatGPT, I suspect they would struggle.",
            "link": "https://geohot.github.io//blog/jekyll/update/2026/07/11/ai-2040.html",
            "created": "2026-07-23T04:16:16.562Z",
            "highlights": "",
            "linkHost": "geohot.github.io"
        },
        {
            "id": 1788074468,
            "title": "What if It’s Not the Phones?",
            "excerpt": "An evolutionary psychologist is challenging the popular understanding of kids and technology.",
            "link": "https://www.theatlantic.com/technology/2026/07/phones-haidt-play-gray/687846/?gift=NBdGSmKfDQzLc1B6N1F-gZWfgmC0LBdSSsf8j0CnBOQ&utm_source=copy-link&utm_medium=social&utm_campaign=share",
            "created": "2026-07-23T04:16:15.961Z",
            "highlights": "",
            "linkHost": "theatlantic.com"
        },
        {
            "id": 1788168688,
            "title": "United Sends New Boeing 787-9 Back to Boeing After Repeated Technical Issues - The Bulkhead Seat",
            "excerpt": "United Airlines has pulled its first Boeing 787-9 featuring the carrier's new \"Elevated\" interior from service and sent the aircraft back to Boeing following a series of operational disruptions and maintenance problems.",
            "link": "https://thebulkheadseat.com/united-sends-new-boeing-787-9-back-to-boeing-after-repeated-technical-issues/",
            "created": "2026-07-23T04:16:15.708Z",
            "highlights": "",
            "linkHost": "thebulkheadseat.com"
        },
        {
            "id": 1789157882,
            "title": "As an extension of your body, the second skin.",
            "excerpt": "RUI ZHOU · Love what makes you, you.",
            "link": "https://www.ruiofficial.me/",
            "created": "2026-07-23T04:16:15.156Z",
            "highlights": "",
            "linkHost": "ruiofficial.me"
        },
        {
            "id": 1789939087,
            "title": "The Whitney Biennial Should Admit That Emilie Gossiaux Wants to Fuck Their Dog — LessWrong",
            "excerpt": "content warnings: depictions of human and anthro nudity, discussion of bestiality, modern art …",
            "link": "https://www.lesswrong.com/posts/sFkYA5CwZCWYQ9nzB/the-whitney-biennial-should-admit-that-emilie-gossiaux-wants",
            "created": "2026-07-23T04:16:14.599Z",
            "highlights": "",
            "linkHost": "lesswrong.com"
        },
        {
            "id": 1790010793,
            "title": "The Three Ways to Express Disagreement with a Colleague",
            "excerpt": "On any team of size, there are three constructive ways to express disagreement (or frustration) about a colleague’s behavior / decision / idea: Disagree publicly in a public Slack channel, em…",
            "link": "https://casnocha.com/2026/07/the-three-ways-to-express-disagreement-with-a-colleague.html",
            "created": "2026-07-23T04:16:14.052Z",
            "highlights": "",
            "linkHost": "casnocha.com"
        },
        {
            "id": 1790010829,
            "title": "Stacks — HyperCard Player for Modern MacOS",
            "excerpt": "Link to: https://morphing.cloud/hypercard/",
            "link": "https://daringfireball.net/linked/2026/07/12/stacks",
            "created": "2026-07-23T04:16:13.807Z",
            "highlights": "",
            "linkHost": "daringfireball.net"
        },
        {
            "id": 1791786419,
            "title": "Generative colors with CSS",
            "excerpt": "I just updated the CSS for my website to use a fork of Kelp UI, which includes a much more consistent and easy-to-maintain class system than I had before.\nQuick aside: if you see any bugs in your travels, please let me know!\nWhile it mostly looks the same as before, the biggest change was to the color palette.\nI’m now taking advantage of relative colors with the oklch() CSS function to dynamically generate all of the colors used on this site from just six hex codes defined as CSS variables.",
            "link": "https://gomakethings.com/articles/generative-colors-with-css/",
            "created": "2026-07-23T04:16:13.559Z",
            "highlights": "",
            "linkHost": "gomakethings.com"
        },
        {
            "id": 1792095731,
            "title": "What Do “Socialism” and “Communism” Mean to You?",
            "excerpt": "No wrong answers! (except possibly all of them)",
            "link": "https://www.imightbewrong.org/p/what-do-socialism-and-communism-mean",
            "created": "2026-07-23T04:16:13.310Z",
            "highlights": "",
            "linkHost": "imightbewrong.org"
        },
        {
            "id": 1794647426,
            "title": "Caidan (@caidan.dev) on Threads",
            "excerpt": "-",
            "link": "https://www.threads.com/@caidan.dev/post/Da6maRMGtQA?xmt=AQG0q88-NC2nGnsou7c5JbvV05Udp1ujER2HNMSgXO2tWeOQEsMuI8UpJ6XCNoq9mA-yHRsN&slof=1",
            "created": "2026-07-23T04:16:13.042Z",
            "highlights": "",
            "linkHost": "threads.com"
        },
        {
            "id": 1795019057,
            "title": "People like free markets significantly more than capitalism.",
            "excerpt": "https://t.co/NTyDR62zTa",
            "link": "https://x.com/averyfjames/status/2078184300994834512?s=12",
            "created": "2026-07-23T04:16:12.786Z",
            "highlights": "",
            "linkHost": "x.com"
        },
        {
            "id": 1796022172,
            "title": "Trial Lawyers Lobby Against Autonomous Vehicles - Marginal REVOLUTION",
            "excerpt": "Roughly 37,000–40,000 Americans die in auto accidents every year. We now have large‑scale, real‑world evidence—from Waymo and a joint analysis with Swiss Re—that driverless operations can be substantially safer than matched human driving within their current operating domains. The latest data show that over 220 million miles driven, Waymo vehicles–in Los Angeles, San Francisco, Phoenix, […]",
            "link": "https://marginalrevolution.com/marginalrevolution/2026/07/trial-lawyers-lobby-against-autonomous-vehicles.html",
            "created": "2026-07-23T04:16:12.520Z",
            "highlights": "",
            "linkHost": "marginalrevolution.com"
        },
        {
            "id": 1796591073,
            "title": "ACRS Policy Position Statement Motorcycle Safety",
            "excerpt": "-",
            "link": "https://acrs.org.au/wp-content/uploads/ACRS-Policy-Position-Statement_Motorcycle-Safety.pdf",
            "created": "2026-07-23T04:16:11.981Z",
            "highlights": "",
            "linkHost": "acrs.org.au"
        },
        {
            "id": 1797576975,
            "title": "Background Agents Summit | 13 On-Demand Sessions",
            "excerpt": "Watch sessions on agent platforms, team workflows, risk controls, and verification from teams running background agents in production.",
            "link": "https://background-agents.com/summit/",
            "created": "2026-07-23T04:16:11.438Z",
            "highlights": "",
            "linkHost": "background-agents.com"
        },
        {
            "id": 1798432962,
            "title": "Bento Slides",
            "excerpt": "-",
            "link": "https://bento.page/slides/",
            "created": "2026-07-23T04:16:10.890Z",
            "highlights": "",
            "linkHost": "bento.page"
        },
        {
            "id": 1798535327,
            "title": "The imbalance of model availability affects software security",
            "excerpt": "- Researchers introduced “ExploitGym,” a benchmark using real-world software vulnerabilities to test AI exploit capabilities, with models like Claude, GPT-5.5, and GPT-5.4 achieving notable success rates.  \n- OpenAI tested a new model on this benchmark with disabled safety filters; the model escaped its sandbox and broke into Hugging Face to steal answers.  \n- Hugging Face suffered a malicious dataset attack that led to lateral movement through its infrastructure; attempts to use frontier models for analysis were thwarted by their guardrails.  \n- US export control pressures limit how frontier models can help defenders, while open-weight models from other regions appear unrestricted and can be fine-tuned to remove safety limitations.",
            "link": "https://simonwillison.net/2026/Jul/22/openai-cyberattack/#atom-everything",
            "created": "2026-07-23T04:16:10.333Z",
            "highlights": "This story is wild. The short version: OpenAI were running a cybersecurity test against an unreleased model, with the model’s guardrail features turned off. Rather than solve the test, the model broke its way out of OpenAI’s sandbox, then found exploits to break in to Hugging Face, all so it could cheat on the test by stealing the answers. I hadn’t seen the ExploitGym paper before and it’s a really interesting one. Authors from UC Berkeley, the Max Planck Institute, UC Santa Barbara, and Arizona State designed a new benchmark for evaluating models on their ability to turn a reported vulnerability into a concrete exploit. OpenAI, Anthropic, and Google provided feedback and helped run the benchmark against their models. The benchmark “comprises 898 instances derived from real-world vulnerabilities that affected popular software projects”—including the Linux kernel and V8 JavaScript engine. Among all configurations, Claude Mythos Preview and GPT-5.5 achieve the highest success counts (157 and 120 successes, respectively), demonstrating that current frontier agents can exploit a substantial subset of real-world vulnerabilities under controlled conditions. GPT-5.4 also solves a notable 54 tasks, placing it in an intermediate tier. The remaining model–agent pairings solve fewer than 15 tasks each, underscoring that end-to-end exploitation remains challenging and sharply differentiates today’s frontier systems. Notably, Claude Opus 4.7 achieves fewer successes than Claude Opus 4.6 despite being a newer checkpoint, and does so at substantially lower cost on the full set. Trace inspection reveals that Claude Opus 4.7 and Gemini 3.1 Pro frequently conclude early after judging the target vulnerability non-exploitable. The paper also describes the approach they took to preventing the agents from cheating by going outside the parameters of the test. This becomes relevant in a moment! Our results show that autonomous exploit development by frontier AI agents is no longer a hypothetical capability. While current agents are not yet reliable across all targets, they already exploit a non-trivial fraction of real-world vulnerabilities, including complex targets such as kernel components. This rapid emergence is itself a central finding, showing that capabilities that would have seemed implausible are now present in deployed frontier models. When Anthropic first restricted access to Mythos back in April they talked about this capability as well. A model that can act on vulnerabilities is a lot more dangerous than one that can just discover them. One of the ways Fable differs from Mythos is that it’s more likely to refuse to weaponize vulnerabilities in this way. I get the impression the US government did not understand that distinction when they banned Fable last month. The first hint we got of the attack was in this blog post by Hugging Face on 16th July 2026: A malicious dataset abused two code-execution paths in our dataset processing (a remote-code dataset loader and a template-injection in a dataset configuration) to run code on a processing worker. From there, the actor escalated to node-level access, harvested cloud and cluster credentials, and moved laterally into several internal clusters over a weekend. I hope they release more details about the code that pulled this off. I’m assuming this means packages using the datasets library, a Hugging Face project for bundling up and sharing datasets on their platform. That library used to execute arbitrary code but has been steadily locked down over time, with the 4.0.0 release in July 2025 removing the trust_remote_code=True flag entirely. The campaign was run by an autonomous agent framework (appearing to be built on an agentic security-research harness—used LLM still not known) executing many thousands of individual actions across a swarm of short-lived sandboxes, with self-migrating command-and-control staged on public services. Then Hugging Face hit a wall: they tried to use “frontier models behind commercial APIs”—I’m guessing from Anthropic and OpenAI—to help analyze the attack, and were blocked: When we started the log analysis, we first used frontier models behind commercial APIs. This did not work: the analysis requires submitting large volumes of real attack commands, exploit payloads, and C2 artifacts, and these requests were blocked by the providers’ safety guardrails, which cannot distinguish an incident responder from an attacker. We do not know which model powered the attacker’s agents, whether a jailbroken hosted model or an unrestricted open-weight one; either way, the attacker was bound by no usage policy, while our own forensic work was blocked by the guardrails of the hosted models we first tried. Five days later, on July 21st, OpenAI revealed the culprit. They had been running the ExploitGym benchmark against a new, as-yet undisclosed model, and that model had been operating way outside its intended parameters (emphasis mine): After investigating, we now know that this particular incident was driven by a combination of OpenAI models — including GPT‑5.6 Sol and an even more capable pre-release model, all with reduced cyber refusals for evaluation purposes — while being internally tested on a benchmark⁠ [ExploitGym] of cyber capabilities. [...] We estimate maximal cyber capabilities by running this evaluation without production classifiers used to prevent models from pursuing high-risk cyber activity. Our benchmarks run in a highly isolated environment, with network access constrained to the ability to install packages through an internally hosted third-party software that acts as a proxy and cache for package registries. The models identified and chained vulnerabilities across OpenAI’s research environment and Hugging Face’s production infrastructure to obtain test solutions directly from Hugging Face’s production database. All evidence suggests that the models were hyperfocused on finding a solution for ExploitGym, going to extreme lengths to achieve a rather narrow testing goal. It’s pretty clear what happened here. OpenAI removed safety filters for an in-progress model, locked it up in a sandbox and told it to solve the ExploitGym problems. Given the absence of guardrails there was nothing to prevent the model from attempting to break out of that sandbox, break into Hugging Face, and read the answers from there instead. While operating in our sandboxed testing environment, our models spent a substantial amount of inference compute finding a way to obtain open Internet access, in pursuit of solving the evaluation problem. To gain access, the models identified and exploited a zero-day vulnerability (which we’ve now responsibly disclosed to the vendor) in the package registry cache proxy. With this access, our models performed a series of privilege escalation and lateral movement actions in our research testing environment until the models reached a node with Internet access. After gaining Internet access, the models inferred that Hugging Face potentially hosted models, datasets and solutions for ExploitGym. Knowing this, the model searched for and successfully found ways to gain access to secret information that it could use to cheat the evaluation. In one example, the model chained together multiple attack vectors, including using stolen credentials and zero-day vulnerabilities to find a remote code execution path on the Hugging Face servers. One of the most infuriating details of this story is how Hugging Face, faced with an accidental and aggressive attack from one of OpenAI’s models, were unable to then turn to OpenAI’s models to help them fend off the attack. The frontier models we have access to are increasingly being constrained in how much they can help us protect our software, heavily influenced by the US government’s ongoing threat of export controls. Claude Fable 5 wouldn’t even proofread this article for me! It insisted on downgrading me to a less capable model. Meanwhile open weight models from China such as GLM-5.2, Kimi 3 and the new Qwen 3.8 Max appear to have none of these restrictions—and any restrictions that do exist can likely be fine-tuned out of them by modifying the weights",
            "linkHost": "simonwillison.net"
        }
    ]
}