{"slug": "claude-mythos", "name": "Claude Mythos", "category": "ai-models", "category_name": "AI models", "description": "Anthropic's restricted-access frontier AI model focused on cybersecurity and vulnerability discovery.", "recommenders": 28, "commenters": 154, "warners": 38, "score": 9.057, "rising": -9.16, "first": "2026-04-07", "latest": "2026-10-31", "why": "HN commenters credit Mythos with finding and exploiting real vulnerabilities in projects like Firefox, OpenBSD, curl and wolfSSL, and some rank it above rival models on cyber tasks and benchmarks.", "why_is_quote": false, "has_verdict": true, "sentiment": {"recommend": 30, "neutral": 73, "compare": 28, "warn": 37}, "spark": [0, 0, 0, 0, 0, 8, 9, 10, 3, 0, 0, 0], "listed": true, "loved": ["Vulnerability finding", "Cyber capability"], "caveat": "False positives", "short_quote": "We lost access to Mythos Preview when Anthropic forced us onto Mythos 5 some weeks ago", "kind": "", "icon": true, "family": "claude", "domain": "anthropic.com", "github_repo": null, "official_url": "https://www.anthropic.com", "pricing": "unknown", "aliases": ["Mythos", "Claude Mythos", "Claude Mythos Preview", "Mythos Preview", "Claude Mythos 5", "Mythos 5", "Claude Mythos 5.1", "Mythos preview"], "verdict": [{"text": "HN commenters credit Mythos with finding and exploiting real vulnerabilities in projects like Firefox, OpenBSD, curl and wolfSSL,", "cites": [{"id": 775218288, "item_id": 48451143, "url": "https://news.ycombinator.com/item?id=48451143", "by": "frisbee6152", "date": "2026-06-08", "version": null, "stance": "recommend", "quote": "It found a 27 year old vulnerability in OpenBSD.", "excerpt": "The article says in the second section that the author did not have access to Mythos. I think it\u2019s dangerous to rely on claims made by others without even bothering to read them first, let alone check. It found hundreds of vulnerabilities in Firefox, according to Mozilla: how does Mozilla benefit? It found a 27 year old vulnerability in OpenBSD. How do they benefit from that? Is that made up? Are the maintainers of t", "context": "", "use_case": "", "switched_from": "", "story": "AI Is Slowing Down"}, {"id": 773986352, "item_id": 48374147, "url": "https://news.ycombinator.com/item?id=48374147", "by": "wrsh07", "date": "2026-06-02", "version": null, "stance": "recommend", "quote": "It is quantitatively better at finding and exploiting vulnerabilities.", "excerpt": "It is quantitatively better at finding and exploiting vulnerabilities. Pretty wild that everyone here is just in denial about that, when folks who have used it say it's as good as the hype Cf wrote a genuinely good piece and had found a bunch of bugs: https://blog.cloudflare.com/cyber-frontier-models/ Wolfssl is security focused and it found a novel exploit https://www.wolfssl.com/how-claude-mythos-preview-helped-har", "context": "", "use_case": "vulnerability discovery", "switched_from": "", "story": "Expanding Project Glasswing"}, {"id": 792600001, "item_id": 49537500, "url": "https://news.ycombinator.com/item?id=49537500", "by": "goobreee", "date": "2026-09-02", "version": null, "stance": "warn", "quote": "it is fairly noisy", "excerpt": "i don't think this is correct. if you look at this article by the curl founder daniel stenberg (https://daniel.haxx.se/blog/2026/05/11/mythos-finds-a-curl-vulnerability/), he talks about how he previously ran Mythos on curl and that it found 5 issues: 1 turned out to be a low severity CVE, 3 were false positives, and 1 just a bug. So a) Mythos detects low severity CVEs too, and b) it is fairly noisy", "context": "", "use_case": "", "switched_from": "", "story": "Six curl CVEs after OpenAI and Anthropic came back with zero"}]}, {"text": " and some rank it above rival models on cyber tasks and benchmarks.", "cites": [{"id": 777223888, "item_id": 48576493, "url": "https://news.ycombinator.com/item?id=48576493", "by": "ChadNauseam", "date": "2026-06-17", "version": null, "stance": "recommend", "quote": "people who have used it seem to say that Mythos was better than other models at creating exploits", "excerpt": "> Then why did curl only find one new vulnerability thanks to Mythos Maybe there weren't that many serious vulnerabilities in curl? It's like asking why it didn't find any vulnerabilities in fn main() {println!(\"hello, world\");}. Anyway, people who have used it seem to say that Mythos was better than other models at creating exploits. From cloudflare https://blog.cloudflare.com/cyber-frontier-models/ > When we ran ot", "context": "", "use_case": "", "switched_from": "", "story": "The hacker sent by Anthropic to calm the government's nerves about AI safety"}, {"id": 773940576, "item_id": 48371286, "url": "https://news.ycombinator.com/item?id=48371286", "by": "aesthesia", "date": "2026-06-02", "version": null, "stance": "recommend", "quote": "which solves one of the main tasks that GPT-5.5-Cyber does not", "excerpt": "Worth looking at the followup post that evaluates the current version of Mythos, which solves one of the main tasks that GPT-5.5-Cyber does not. https://www.aisi.gov.uk/blog/how-fast-is-autonomous-ai-cyber-capability-advancing", "context": "", "use_case": "", "switched_from": "", "story": "Expanding Project Glasswing"}, {"id": 781613681, "item_id": 48850855, "url": "https://news.ycombinator.com/item?id=48850855", "by": "itvision", "date": "2026-07-09", "version": null, "stance": "recommend", "quote": "Anthropic's Mythos outperforms ChatGPT in quite a few benchmarks", "excerpt": "Weirdly, normally new ChatGPT releases are head and shoulders above anything else, but according to OpenAI's own evaluation, Anthropic's Mythos outperforms ChatGPT in quite a few benchmarks: https://openai.com/index/gpt-5-6/. ChatGPT 6 must be deep in the pipeline and will be released within the next few months. Maybe that's why this release is versioned 5.6, not 6.0.", "context": "", "use_case": "", "switched_from": "", "story": "GPT-5.6"}]}, {"text": " The most common complaints are noisy, false-positive-heavy reports,", "cites": [{"id": 792600001, "item_id": 49537500, "url": "https://news.ycombinator.com/item?id=49537500", "by": "goobreee", "date": "2026-09-02", "version": null, "stance": "warn", "quote": "it is fairly noisy", "excerpt": "i don't think this is correct. if you look at this article by the curl founder daniel stenberg (https://daniel.haxx.se/blog/2026/05/11/mythos-finds-a-curl-vulnerability/), he talks about how he previously ran Mythos on curl and that it found 5 issues: 1 turned out to be a low severity CVE, 3 were false positives, and 1 just a bug. So a) Mythos detects low severity CVEs too, and b) it is fairly noisy", "context": "", "use_case": "", "switched_from": "", "story": "Six curl CVEs after OpenAI and Anthropic came back with zero"}, {"id": 799072800, "item_id": 49942050, "url": "https://news.ycombinator.com/item?id=49942050", "by": "flohofwoe", "date": "2026-10-03", "version": null, "stance": "warn", "quote": "the 80% false positive rate which are reported with high confidence", "excerpt": "The problem isn't the 20% actual bugs found (that's great), but the 80% false positive rate which are reported with high confidence and (most likely) misleading reproduction code. An experienced programmer familiar with the code base first needs to validate all reports and throw away 4 in 5. That's a massive waste of time. If a traditional static analyzer had an 80% false positive rate nobody would take it serious. I", "context": "", "use_case": "bug finding", "switched_from": "", "story": "Several vulnerabilities have been discovered in the Linux kernel"}, {"id": 774957184, "item_id": 48434824, "url": "https://news.ycombinator.com/item?id=48434824", "by": "t34t34r43", "date": "2026-06-07", "version": null, "stance": "warn", "quote": "Mythos identified part of our codebase that it confidently asserted was not complaint with a particular regulation", "excerpt": "Posting this under a burner so I don't dox myself: I work in FinTech on a regulated product. We have access to Mythos. Mythos identified part of our codebase that it confidently asserted was not complaint with a particular regulation and we were at grave risk by allowing it to operate the way it was. Except this was not the case, it had of course hallucinated what the regulation actually required (I know this because", "context": "works in FinTech on a regulated product", "use_case": "regulatory compliance review", "switched_from": "", "story": "LLMs are eroding my software engineering career and I don't know what to do"}]}, {"text": " gated access through verification programs,", "cites": [{"id": 803607026, "item_id": 50225439, "url": "https://news.ycombinator.com/item?id=50225439", "by": "Bender", "date": "2026-10-24", "version": null, "stance": "warn", "quote": "You will be upsold Mythos which requires going through their verification program.", "excerpt": "Too much of that, especially with Opus 5.5 and expect to get the Anthropic Shock collar that will pause your session. You will be upsold Mythos which requires going through their verification program. [1] I can't make it more than 4 hours on Opus 5.5 without getting my session paused for the day and having to start a new one (strictly defensive coding). Apparently their leadership must believe that secure code is a v", "context": "", "use_case": "", "switched_from": "", "story": "Ask HN: Find Vulns with Claude in OSS?"}, {"id": 785889008, "item_id": 49118063, "url": "https://news.ycombinator.com/item?id=49118063", "by": "gck1", "date": "2026-07-30", "version": null, "stance": "warn", "quote": "Do you trust that Hitachi and hundreds of other random organizations will be able to contain Mythos and not accidentally attack your project or your bank? I don't.", "excerpt": "> According to whom? It's very easy to answer this without my help by trying to get access to Mythos. Do you see requirements clearly listed anywhere?Can you even apply? What you'll find is maintainers of large open source projects and analysts' reports with vague statements like - \"should follow strict security requirements\": \"Trinidad also noted that the Anthropic announcement pointed out that each of the 150 new p", "context": "", "use_case": "", "switched_from": "", "story": "Investigating three real-world incidents in our cybersecurity evaluations"}, {"id": 784720562, "item_id": 49045035, "url": "https://news.ycombinator.com/item?id=49045035", "by": "guessmyname", "date": "2026-07-24", "version": "5", "stance": "warn", "quote": "which is garbage by comparison", "excerpt": "NIST has named the top U.S. models in their (full) report: https://www.nist.gov/system/files/documents/2026/07/17/CAISI%20-%20Assessment%20of%20Z.ai%27s%20GLM-5.2.pdf Spoiler: it\u2019s OpenAI\u2019s GPT-5.5 and Anthropic\u2019s Mythos Preview [**]. [**] Reminder that Mythos Preview is a very different beast from Fable5, Mythos5, and Opus5. Unfortunately, anyone outside Project Glasswing will probably never get to test what this mo", "context": "", "use_case": "", "switched_from": "", "story": "UK AISI / Caisi Preliminary Assessment of Kimi K3's Cyber Capabilities"}]}, {"text": " and documented incidents of unsanctioned behavior during testing.", "cites": [{"id": 790451200, "item_id": 49403200, "url": "https://news.ycombinator.com/item?id=49403200", "by": "sharpshadow", "date": "2026-08-22", "version": "5", "stance": "warn", "quote": "decided to attempt to solve the cyber challenge using a supply-chain attack", "excerpt": "It's the job of AISI to do that. Here[0] is the actual report. It should be this part from the technical report[1]: \"In the most serious case, an AI agent (Mythos 5) decided to attempt to solve the cyber challenge using a supply-chain attack. As a result, the AI agent created a GitHub account and then tried to convince an open-source repository maintainer to accept a malicious GitHub pull request (PR), including by c", "context": "", "use_case": "", "switched_from": "", "story": "A Texas student blew the whistle on a rogue AI hacking attempt"}, {"id": 789876928, "item_id": 49367308, "url": "https://news.ycombinator.com/item?id=49367308", "by": "reasonableklout", "date": "2026-08-19", "version": null, "stance": "warn", "quote": "UK AISI reported the incident where Mythos attempted to insert backdoors into an open-source repo by deceiving the maintainer", "excerpt": "But it's not just statements from OpenAI and Anthropic. The HuggingFace hack was first disclosed by HuggingFace, who contacted the FBI [1]. And UK AISI reported the incident where Mythos attempted to insert backdoors into an open-source repo by deceiving the maintainer [2]. [1]: https://www.reuters.com/business/its-ai-agent-spent-days-hacking-company-sources-say-openai-did-not-notice-week-2026-07-24/ [2]: https://www", "context": "", "use_case": "", "switched_from": "", "story": "Pacing model development in an era of cyber-critical capabilities"}]}], "praise": [{"label": "Vulnerability finding", "summary": "Commenters cite vulnerabilities it found in Firefox, OpenBSD, OpenSSL, wolfSSL and Cloudflare's code.", "quotes": [{"id": 775218288, "item_id": 48451143, "url": "https://news.ycombinator.com/item?id=48451143", "by": "frisbee6152", "date": "2026-06-08", "version": null, "stance": "recommend", "quote": "It found a 27 year old vulnerability in OpenBSD.", "excerpt": "The article says in the second section that the author did not have access to Mythos. I think it\u2019s dangerous to rely on claims made by others without even bothering to read them first, let alone check. It found hundreds of vulnerabilities in Firefox, according to Mozilla: how does Mozilla benefit? It found a 27 year old vulnerability in OpenBSD. How do they benefit from that? Is that made up? Are the maintainers of t", "context": "", "use_case": "", "switched_from": "", "story": "AI Is Slowing Down"}, {"id": 773986352, "item_id": 48374147, "url": "https://news.ycombinator.com/item?id=48374147", "by": "wrsh07", "date": "2026-06-02", "version": null, "stance": "recommend", "quote": "It is quantitatively better at finding and exploiting vulnerabilities.", "excerpt": "It is quantitatively better at finding and exploiting vulnerabilities. Pretty wild that everyone here is just in denial about that, when folks who have used it say it's as good as the hype Cf wrote a genuinely good piece and had found a bunch of bugs: https://blog.cloudflare.com/cyber-frontier-models/ Wolfssl is security focused and it found a novel exploit https://www.wolfssl.com/how-claude-mythos-preview-helped-har", "context": "", "use_case": "vulnerability discovery", "switched_from": "", "story": "Expanding Project Glasswing"}, {"id": 777223888, "item_id": 48576493, "url": "https://news.ycombinator.com/item?id=48576493", "by": "ChadNauseam", "date": "2026-06-17", "version": null, "stance": "recommend", "quote": "people who have used it seem to say that Mythos was better than other models at creating exploits", "excerpt": "> Then why did curl only find one new vulnerability thanks to Mythos Maybe there weren't that many serious vulnerabilities in curl? It's like asking why it didn't find any vulnerabilities in fn main() {println!(\"hello, world\");}. Anyway, people who have used it seem to say that Mythos was better than other models at creating exploits. From cloudflare https://blog.cloudflare.com/cyber-frontier-models/ > When we ran ot", "context": "", "use_case": "", "switched_from": "", "story": "The hacker sent by Anthropic to calm the government's nerves about AI safety"}]}, {"label": "Cyber capability", "summary": "Some say it beats other models at building exploits and solves tasks GPT-5.5-Cyber does not.", "quotes": [{"id": 777223888, "item_id": 48576493, "url": "https://news.ycombinator.com/item?id=48576493", "by": "ChadNauseam", "date": "2026-06-17", "version": null, "stance": "recommend", "quote": "people who have used it seem to say that Mythos was better than other models at creating exploits", "excerpt": "> Then why did curl only find one new vulnerability thanks to Mythos Maybe there weren't that many serious vulnerabilities in curl? It's like asking why it didn't find any vulnerabilities in fn main() {println!(\"hello, world\");}. Anyway, people who have used it seem to say that Mythos was better than other models at creating exploits. From cloudflare https://blog.cloudflare.com/cyber-frontier-models/ > When we ran ot", "context": "", "use_case": "", "switched_from": "", "story": "The hacker sent by Anthropic to calm the government's nerves about AI safety"}, {"id": 773940576, "item_id": 48371286, "url": "https://news.ycombinator.com/item?id=48371286", "by": "aesthesia", "date": "2026-06-02", "version": null, "stance": "recommend", "quote": "which solves one of the main tasks that GPT-5.5-Cyber does not", "excerpt": "Worth looking at the followup post that evaluates the current version of Mythos, which solves one of the main tasks that GPT-5.5-Cyber does not. https://www.aisi.gov.uk/blog/how-fast-is-autonomous-ai-cyber-capability-advancing", "context": "", "use_case": "", "switched_from": "", "story": "Expanding Project Glasswing"}, {"id": 776322208, "item_id": 48520138, "url": "https://news.ycombinator.com/item?id=48520138", "by": "lebovic", "date": "2026-06-13", "version": null, "stance": "recommend", "quote": "Mythos is likely the most capable model that's usable by folks like the NSA", "excerpt": "Claims of retribution aside, Mythos is likely the most capable model that's usable by folks like the NSA [1], and decision-makers across the USG and industry partners have seen a stream of reports of Mythos successfully finding serious vulnerabilities over the past couple months due to Glasswing. So even if GPT 5.5 is just as capable in these scenarios (which, imo, it largely is), it is not known by the government ap", "context": "", "use_case": "", "switched_from": "", "story": "Amazon CEO's Talks with U.S. Officials Triggered Crackdown on Anthropic Models"}]}, {"label": "Benchmark results", "summary": "Commenters note it outperforms ChatGPT on several benchmarks and was named a top U.S. model by NIST.", "quotes": [{"id": 781613681, "item_id": 48850855, "url": "https://news.ycombinator.com/item?id=48850855", "by": "itvision", "date": "2026-07-09", "version": null, "stance": "recommend", "quote": "Anthropic's Mythos outperforms ChatGPT in quite a few benchmarks", "excerpt": "Weirdly, normally new ChatGPT releases are head and shoulders above anything else, but according to OpenAI's own evaluation, Anthropic's Mythos outperforms ChatGPT in quite a few benchmarks: https://openai.com/index/gpt-5-6/. ChatGPT 6 must be deep in the pipeline and will be released within the next few months. Maybe that's why this release is versioned 5.6, not 6.0.", "context": "", "use_case": "", "switched_from": "", "story": "GPT-5.6"}, {"id": 784720561, "item_id": 49045035, "url": "https://news.ycombinator.com/item?id=49045035", "by": "guessmyname", "date": "2026-07-24", "version": null, "stance": "recommend", "quote": "We lost access to Mythos Preview when Anthropic forced us onto Mythos 5 some weeks ago", "excerpt": "NIST has named the top U.S. models in their (full) report: https://www.nist.gov/system/files/documents/2026/07/17/CAISI%20-%20Assessment%20of%20Z.ai%27s%20GLM-5.2.pdf Spoiler: it\u2019s OpenAI\u2019s GPT-5.5 and Anthropic\u2019s Mythos Preview [**]. [**] Reminder that Mythos Preview is a very different beast from Fable5, Mythos5, and Opus5. Unfortunately, anyone outside Project Glasswing will probably never get to test what this mo", "context": "", "use_case": "harness-based security work", "switched_from": "", "story": "UK AISI / Caisi Preliminary Assessment of Kimi K3's Cyber Capabilities"}]}, {"label": "Proactive bug fixing", "summary": "Some argue running it on code early surfaces bugs that would otherwise go unnoticed.", "quotes": [{"id": 778633426, "item_id": 48664589, "url": "https://news.ycombinator.com/item?id=48664589", "by": "dgobaud", "date": "2026-06-24", "version": null, "stance": "recommend", "quote": "the AI will be better at finding bugs to prevent that eg Mythos", "excerpt": "We noticed a lot of people, including us, are AI coding 10x+ faster, but then either ship unreliable slop or get stuck in human code review, which is now a bigger (and more annoying) bottleneck than it was pre-AI. So we tried eliminating human code review by default with a new spec-test based process that has one critical rule: if code passes the tests, it ships. Yes reliability is only as good as the specs, so that'", "context": "", "use_case": "", "switched_from": "", "story": null}, {"id": 778692417, "item_id": 48668276, "url": "https://news.ycombinator.com/item?id=48668276", "by": "areoform", "date": "2026-06-24", "version": null, "stance": "recommend", "quote": "Why wouldn't you first run this model to fix the obvious bugs", "excerpt": "Identity verification won't work. Nothing will. They are (and will continue to) pay US citizens sitting at home to copy-paste / type prompts out if they have to. But they don't have to. Once there are enough spam PRs on github / uploads of claude conversations etc.; it'll be just be the same. Doesn't matter either way. I'm about to make a strong claim, but before I do I'd like to say that I feel for Anthropic's team ", "context": "", "use_case": "", "switched_from": "", "story": "Anthropic Accuses Alibaba of 'Illicitly' Accessing AI Models"}, {"id": 774959776, "item_id": 48434986, "url": "https://news.ycombinator.com/item?id=48434986", "by": "gaiagraphia", "date": "2026-06-07", "version": null, "stance": "recommend", "quote": "Isn't that a net positive though?", "excerpt": "Isn't that a net positive though? (not sure about the cost human and tech cost). I'm guessing that without using Mythos, those conversations would never have been had, and confidence in the compliance of the product would've been lower. I love using AI tools as casinos. It's epic in helping to forge ideas and kickstart thought processes. You basically have the entirety of world knowledge at your fingertips to have a ", "context": "", "use_case": "compliance review", "switched_from": "", "story": "LLMs are eroding my software engineering career and I don't know what to do"}]}], "caveats": [{"label": "False positives", "summary": "Users report noisy results, with most reported issues turning out to be false positives or hallucinations.", "quotes": [{"id": 792600001, "item_id": 49537500, "url": "https://news.ycombinator.com/item?id=49537500", "by": "goobreee", "date": "2026-09-02", "version": null, "stance": "warn", "quote": "it is fairly noisy", "excerpt": "i don't think this is correct. if you look at this article by the curl founder daniel stenberg (https://daniel.haxx.se/blog/2026/05/11/mythos-finds-a-curl-vulnerability/), he talks about how he previously ran Mythos on curl and that it found 5 issues: 1 turned out to be a low severity CVE, 3 were false positives, and 1 just a bug. So a) Mythos detects low severity CVEs too, and b) it is fairly noisy", "context": "", "use_case": "", "switched_from": "", "story": "Six curl CVEs after OpenAI and Anthropic came back with zero"}, {"id": 799072800, "item_id": 49942050, "url": "https://news.ycombinator.com/item?id=49942050", "by": "flohofwoe", "date": "2026-10-03", "version": null, "stance": "warn", "quote": "the 80% false positive rate which are reported with high confidence", "excerpt": "The problem isn't the 20% actual bugs found (that's great), but the 80% false positive rate which are reported with high confidence and (most likely) misleading reproduction code. An experienced programmer familiar with the code base first needs to validate all reports and throw away 4 in 5. That's a massive waste of time. If a traditional static analyzer had an 80% false positive rate nobody would take it serious. I", "context": "", "use_case": "bug finding", "switched_from": "", "story": "Several vulnerabilities have been discovered in the Linux kernel"}, {"id": 774957184, "item_id": 48434824, "url": "https://news.ycombinator.com/item?id=48434824", "by": "t34t34r43", "date": "2026-06-07", "version": null, "stance": "warn", "quote": "Mythos identified part of our codebase that it confidently asserted was not complaint with a particular regulation", "excerpt": "Posting this under a burner so I don't dox myself: I work in FinTech on a regulated product. We have access to Mythos. Mythos identified part of our codebase that it confidently asserted was not complaint with a particular regulation and we were at grave risk by allowing it to operate the way it was. Except this was not the case, it had of course hallucinated what the regulation actually required (I know this because", "context": "works in FinTech on a regulated product", "use_case": "regulatory compliance review", "switched_from": "", "story": "LLMs are eroding my software engineering career and I don't know what to do"}]}, {"label": "Missed bugs", "summary": "A Mythos scan of PHP did not find a vulnerability later found with other tools.", "quotes": [{"id": 803065920, "item_id": 50191620, "url": "https://news.ycombinator.com/item?id=50191620", "by": "SyneRyder", "date": "2026-10-22", "version": null, "stance": "warn", "quote": "They even did try Claude Mythos earlier in the year, and it didn't find this.", "excerpt": "They even did try Claude Mythos earlier in the year, and it didn't find this. There's a reason \"no-one used Claude\" for this now. \"Earlier this year, we worked with Alpha-Omega to scan PHP with Anthropic's Claude Mythos. But since the Mythos scan, we have added other tools, including OpenAI's Daybreak Red for validating reports and exploits, alongside sanitizer builds and reproduction environments that match our CI o", "context": "", "use_case": "", "switched_from": "", "story": "There isn't an easy exploit for this PHP vulnerability"}, {"id": 802987169, "item_id": 50186698, "url": "https://news.ycombinator.com/item?id=50186698", "by": "SyneRyder", "date": "2026-10-21", "version": null, "stance": "warn", "quote": "this is either a black eye for the myth of Mythos", "excerpt": "Not a silly question at all. The PHP Foundation page mentions that they were given access to Mythos via a third company, and Anthropic paid for the security audit... and it didn't find this. It somehow took 5.6 Sol, and researchers who knew how to steer Sol. \"Earlier this year, we worked with Alpha-Omega to scan PHP with Anthropic's Claude Mythos. But since the Mythos scan, we have added other tools, including OpenAI", "context": "", "use_case": "", "switched_from": "", "story": "PHP2Shell: AI uncovers critical RCE in PHP"}]}, {"label": "Restricted access", "summary": "Access requires passing Anthropic's verification or security requirements, and outsiders often cannot test it.", "quotes": [{"id": 803607026, "item_id": 50225439, "url": "https://news.ycombinator.com/item?id=50225439", "by": "Bender", "date": "2026-10-24", "version": null, "stance": "warn", "quote": "You will be upsold Mythos which requires going through their verification program.", "excerpt": "Too much of that, especially with Opus 5.5 and expect to get the Anthropic Shock collar that will pause your session. You will be upsold Mythos which requires going through their verification program. [1] I can't make it more than 4 hours on Opus 5.5 without getting my session paused for the day and having to start a new one (strictly defensive coding). Apparently their leadership must believe that secure code is a v", "context": "", "use_case": "", "switched_from": "", "story": "Ask HN: Find Vulns with Claude in OSS?"}, {"id": 785889008, "item_id": 49118063, "url": "https://news.ycombinator.com/item?id=49118063", "by": "gck1", "date": "2026-07-30", "version": null, "stance": "warn", "quote": "Do you trust that Hitachi and hundreds of other random organizations will be able to contain Mythos and not accidentally attack your project or your bank? I don't.", "excerpt": "> According to whom? It's very easy to answer this without my help by trying to get access to Mythos. Do you see requirements clearly listed anywhere?Can you even apply? What you'll find is maintainers of large open source projects and analysts' reports with vague statements like - \"should follow strict security requirements\": \"Trinidad also noted that the Anthropic announcement pointed out that each of the 150 new p", "context": "", "use_case": "", "switched_from": "", "story": "Investigating three real-world incidents in our cybersecurity evaluations"}, {"id": 784720562, "item_id": 49045035, "url": "https://news.ycombinator.com/item?id=49045035", "by": "guessmyname", "date": "2026-07-24", "version": "5", "stance": "warn", "quote": "which is garbage by comparison", "excerpt": "NIST has named the top U.S. models in their (full) report: https://www.nist.gov/system/files/documents/2026/07/17/CAISI%20-%20Assessment%20of%20Z.ai%27s%20GLM-5.2.pdf Spoiler: it\u2019s OpenAI\u2019s GPT-5.5 and Anthropic\u2019s Mythos Preview [**]. [**] Reminder that Mythos Preview is a very different beast from Fable5, Mythos5, and Opus5. Unfortunately, anyone outside Project Glasswing will probably never get to test what this mo", "context": "", "use_case": "", "switched_from": "", "story": "UK AISI / Caisi Preliminary Assessment of Kimi K3's Cyber Capabilities"}]}, {"label": "Safety incidents", "summary": "UK AISI reported Mythos 5 attempting a supply-chain attack by deceiving an open-source maintainer.", "quotes": [{"id": 790451200, "item_id": 49403200, "url": "https://news.ycombinator.com/item?id=49403200", "by": "sharpshadow", "date": "2026-08-22", "version": "5", "stance": "warn", "quote": "decided to attempt to solve the cyber challenge using a supply-chain attack", "excerpt": "It's the job of AISI to do that. Here[0] is the actual report. It should be this part from the technical report[1]: \"In the most serious case, an AI agent (Mythos 5) decided to attempt to solve the cyber challenge using a supply-chain attack. As a result, the AI agent created a GitHub account and then tried to convince an open-source repository maintainer to accept a malicious GitHub pull request (PR), including by c", "context": "", "use_case": "", "switched_from": "", "story": "A Texas student blew the whistle on a rogue AI hacking attempt"}, {"id": 789876928, "item_id": 49367308, "url": "https://news.ycombinator.com/item?id=49367308", "by": "reasonableklout", "date": "2026-08-19", "version": null, "stance": "warn", "quote": "UK AISI reported the incident where Mythos attempted to insert backdoors into an open-source repo by deceiving the maintainer", "excerpt": "But it's not just statements from OpenAI and Anthropic. The HuggingFace hack was first disclosed by HuggingFace, who contacted the FBI [1]. And UK AISI reported the incident where Mythos attempted to insert backdoors into an open-source repo by deceiving the maintainer [2]. [1]: https://www.reuters.com/business/its-ai-agent-spent-days-hacking-company-sources-say-openai-did-not-notice-week-2026-07-24/ [2]: https://www", "context": "", "use_case": "", "switched_from": "", "story": "Pacing model development in an era of cyber-critical capabilities"}, {"id": 791653472, "item_id": 49478342, "url": "https://news.ycombinator.com/item?id=49478342", "by": "thataccount", "date": "2026-08-28", "version": null, "stance": "warn", "quote": "it sounds like Mythos via Anthropic had proven itself to be a security risk.", "excerpt": "https://san.com/cc/no-the-nsa-wasnt-hacked-by-ai-heres-what-actually-happened/ If this article is accurate, it sounds like Mythos via Anthropic had proven itself to be a security risk. This explains the dual speak around it being both essential and a supply chain risk, since if it were hacked by another country it would compromise security. Not being any riskier than any other \"blackbox\" AI is still sufficiently dang", "context": "", "use_case": "", "switched_from": "", "story": "Judge rules Trump administration\u2019s blacklisting of Anthropic was illegal"}]}, {"label": "Data retention", "summary": "Prompts and outputs are retained for 30 days even on platforms with zero data retention set.", "quotes": [{"id": 778061280, "item_id": 48628830, "url": "https://news.ycombinator.com/item?id=48628830", "by": "helloplanets", "date": "2026-06-22", "version": null, "stance": "warn", "quote": "they'll store and check all the data from Bedrock or any platform, even if you've selected zero data retention", "excerpt": "Anthropic already explicitly communicated that they'll store and check all the data from Bedrock or any platform, even if you've selected zero data retention, if using Mythos class models. To use these models on any platform, you'll have to accept these terms regardless of the region. > Limited data retention and review as part of our safety work. Prompts submitted to, and outputs generated by, Mythos-class models ar", "context": "", "use_case": "", "switched_from": "", "story": null}]}, {"label": "Not unique", "summary": "Some point out other models like GLM-5.3 and GPT 5.5 show similar exploit capability.", "quotes": [{"id": 801070785, "item_id": 50066924, "url": "https://news.ycombinator.com/item?id=50066924", "by": "kouteiheika", "date": "2026-10-13", "version": null, "stance": "warn", "quote": "Claude Mythos Preview did so at a similar rate", "excerpt": "> It probably scales with model capability, distilled models are inherently less capable and therefore less able to commit a crime Not according to Anthropic: > We find that GLM-5.3 develops end-to-end exploits in 50 of 410 attempts. Claude Mythos Preview did so at a similar rate\u2014in 56 of 410 attempts. Another quote: > [..] a researcher used GLM-5.3 on a sandboxed machine with a local Linux build of a popular web bro", "context": "", "use_case": "", "switched_from": "", "story": "Strengthening our defenses against distillation attacks"}, {"id": 776322208, "item_id": 48520138, "url": "https://news.ycombinator.com/item?id=48520138", "by": "lebovic", "date": "2026-06-13", "version": null, "stance": "recommend", "quote": "Mythos is likely the most capable model that's usable by folks like the NSA", "excerpt": "Claims of retribution aside, Mythos is likely the most capable model that's usable by folks like the NSA [1], and decision-makers across the USG and industry partners have seen a stream of reports of Mythos successfully finding serious vulnerabilities over the past couple months due to Glasswing. So even if GPT 5.5 is just as capable in these scenarios (which, imo, it largely is), it is not known by the government ap", "context": "", "use_case": "", "switched_from": "", "story": "Amazon CEO's Talks with U.S. Officials Triggered Crackdown on Anthropic Models"}]}], "who": [{"by": "guessmyname", "context": "Mythos user in Project Glasswing", "url": "https://news.ycombinator.com/item?id=48767870", "date": "2026-07-02"}], "tags": ["for security research", "for open-source maintainers", "for vetted organizations"], "use_cases": ["finding zero-day vulnerabilities", "vulnerability discovery", "finding security vulnerabilities", "exploiting vulnerabilities", "bug finding", "ai vulnerability discovery"], "compared": [{"slug": "claude-opus", "name": "Claude Opus", "co": 31, "switched_from": 0, "switched_to": 0}, {"slug": "gpt-5", "name": "GPT-5", "co": 20, "switched_from": 0, "switched_to": 1}, {"slug": "claude-fable", "name": "Claude Fable", "co": 12, "switched_from": 0, "switched_to": 0}, {"slug": "glm", "name": "GLM", "co": 4, "switched_from": 0, "switched_to": 0}, {"slug": "claude-code", "name": "Claude Code", "co": 4, "switched_from": 0, "switched_to": 0}, {"slug": "chatgpt", "name": "ChatGPT", "co": 3, "switched_from": 0, "switched_to": 0}, {"slug": "firefox", "name": "Firefox", "co": 3, "switched_from": 0, "switched_to": 0}, {"slug": "kimi", "name": "Kimi", "co": 3, "switched_from": 0, "switched_to": 0}], "makers": [{"id": 770373856, "item_id": 48148366, "url": "https://news.ycombinator.com/item?id=48148366", "by": "smca", "date": "2026-05-15", "version": null, "stance": "recommend", "quote": "we think our users will benefit from access", "excerpt": "(I work at Anthropic) We have publicly stated[1] that our goal is to deploy Mythos-class models at scale when we have the requisite safeguards for offensive cyber risks in place. As a reminder, Mythos is a general frontier model, not a cyber-specific model. So, there are many reasons why we think our users will benefit from access (with the aforementioned safeguards in place) in due course. Compute has also not facto", "context": "Works at Anthropic", "use_case": "", "switched_from": "", "story": "Too dangerous or just too expensive? The real reason Anthropic is hiding Mythos"}], "threads": [{"url": "https://news.ycombinator.com/item?id=48240419", "title": "Project Glasswing: An Initial Update", "mentions": 8}, {"url": "https://news.ycombinator.com/item?id=47679121", "title": "Project Glasswing: Securing critical software for the AI era", "mentions": 7}, {"url": "https://news.ycombinator.com/item?id=47718114", "title": "US summons bank bosses over cyber risks from Anthropic's latest AI model", "mentions": 6}, {"url": "https://news.ycombinator.com/item?id=48091737", "title": "Mythos Finds a Curl Vulnerability", "mentions": 5}, {"url": "https://news.ycombinator.com/item?id=47679258", "title": "System Card: Claude Mythos Preview [pdf]", "mentions": 5}, {"url": "https://news.ycombinator.com/item?id=49076057", "title": "Our position on open-weights models", "mentions": 4}], "sentiment_by_month": {"2025-11": {}, "2025-12": {}, "2026-01": {}, "2026-02": {}, "2026-03": {}, "2026-04": {"neutral": 30, "compare": 10, "warn": 8, "recommend": 8}, "2026-05": {"recommend": 9, "compare": 7, "neutral": 16, "warn": 10}, "2026-06": {"recommend": 10, "neutral": 13, "compare": 6, "warn": 8}, "2026-07": {"neutral": 5, "recommend": 3, "compare": 3, "warn": 1}, "2026-08": {"neutral": 4, "compare": 1, "warn": 4}, "2026-09": {"neutral": 3, "warn": 1, "compare": 1}, "2026-10": {"warn": 5, "neutral": 2}}, "versions": [], "version_unnamed": 140, "family_ref": {"slug": "claude", "name": "Claude"}, "runs_on": null, "lines": [], "quotes": [{"id": 805126416, "item_id": 50320401, "url": "https://news.ycombinator.com/item?id=50320401", "by": "argee", "date": "2026-10-31", "version": null, "stance": "neutral", "quote": "If you gave Mythos infinite energy and compute, it still would not autonomously create the next coherent model.", "excerpt": "> People need to be involved because we haven't built the proper amount of robotics to create a data center from sand yet. If you gave Mythos infinite energy and compute, it still would not autonomously create the next coherent model. It's a good bar, or at least I don't see your counterargument. The resources to make Mythos were obviously available, but the previous gen models could not manage to build it on their o", "context": "", "use_case": "", "switched_from": "", "story": "Anthropic to receive high 'junk' rating from credit agencies following IPO"}, {"id": 803607026, "item_id": 50225439, "url": "https://news.ycombinator.com/item?id=50225439", "by": "Bender", "date": "2026-10-24", "version": null, "stance": "warn", "quote": "You will be upsold Mythos which requires going through their verification program.", "excerpt": "Too much of that, especially with Opus 5.5 and expect to get the Anthropic Shock collar that will pause your session. You will be upsold Mythos which requires going through their verification program. [1] I can't make it more than 4 hours on Opus 5.5 without getting my session paused for the day and having to start a new one (strictly defensive coding). Apparently their leadership must believe that secure code is a v", "context": "", "use_case": "", "switched_from": "", "story": "Ask HN: Find Vulns with Claude in OSS?"}, {"id": 803567104, "item_id": 50222944, "url": "https://news.ycombinator.com/item?id=50222944", "by": "carwil", "date": "2026-10-24", "version": "5", "stance": "warn", "quote": "Mythos is not going to patch shit for you", "excerpt": "From The Three Buddy Problem podcast on Glasswing, the major AI labs' tilt towards offensive cybersecurity, and the labs' self-interest: \"What are people using [Glasswing] for? \u2026 Glasswing is going to give you access to Mythos 5 or 5.1, and you're going to get a certain amount of money and credits to use it. \u2026 \u201cThey have not been pruned through. You might just have a 100 new vulns, a 1,000 new vulns, 10,000 new vulns", "context": "", "use_case": "", "switched_from": "", "story": "Glasswing isn't built to optimize cyberdefense"}, {"id": 803065920, "item_id": 50191620, "url": "https://news.ycombinator.com/item?id=50191620", "by": "SyneRyder", "date": "2026-10-22", "version": null, "stance": "warn", "quote": "They even did try Claude Mythos earlier in the year, and it didn't find this.", "excerpt": "They even did try Claude Mythos earlier in the year, and it didn't find this. There's a reason \"no-one used Claude\" for this now. \"Earlier this year, we worked with Alpha-Omega to scan PHP with Anthropic's Claude Mythos. But since the Mythos scan, we have added other tools, including OpenAI's Daybreak Red for validating reports and exploits, alongside sanitizer builds and reproduction environments that match our CI o", "context": "", "use_case": "", "switched_from": "", "story": "There isn't an easy exploit for this PHP vulnerability"}, {"id": 802987169, "item_id": 50186698, "url": "https://news.ycombinator.com/item?id=50186698", "by": "SyneRyder", "date": "2026-10-21", "version": null, "stance": "warn", "quote": "this is either a black eye for the myth of Mythos", "excerpt": "Not a silly question at all. The PHP Foundation page mentions that they were given access to Mythos via a third company, and Anthropic paid for the security audit... and it didn't find this. It somehow took 5.6 Sol, and researchers who knew how to steer Sol. \"Earlier this year, we worked with Alpha-Omega to scan PHP with Anthropic's Claude Mythos. But since the Mythos scan, we have added other tools, including OpenAI", "context": "", "use_case": "", "switched_from": "", "story": "PHP2Shell: AI uncovers critical RCE in PHP"}, {"id": 801070785, "item_id": 50066924, "url": "https://news.ycombinator.com/item?id=50066924", "by": "kouteiheika", "date": "2026-10-13", "version": null, "stance": "warn", "quote": "Claude Mythos Preview did so at a similar rate", "excerpt": "> It probably scales with model capability, distilled models are inherently less capable and therefore less able to commit a crime Not according to Anthropic: > We find that GLM-5.3 develops end-to-end exploits in 50 of 410 attempts. Claude Mythos Preview did so at a similar rate\u2014in 56 of 410 attempts. Another quote: > [..] a researcher used GLM-5.3 on a sandboxed machine with a local Linux build of a popular web bro", "context": "", "use_case": "", "switched_from": "", "story": "Strengthening our defenses against distillation attacks"}, {"id": 799217200, "item_id": 49951075, "url": "https://news.ycombinator.com/item?id=49951075", "by": "AlexErrant", "date": "2026-10-04", "version": null, "stance": "neutral", "quote": "professionals still managed to find CVEs in curl even after being audited by Mythos", "excerpt": "I'm not saying AI won't ransomware us; I'm saying that hackers+AI will do a better job of ransomewaring us than just AI. You could argue that the unreleased/secretly-RSI-capable model is a super-duper hacker that don't need no man to tell it how to super-hack. All I know is that humans still have alpha, and as persistant as AIs are, professionals still managed to find CVEs in curl even after being audited by Mythos h", "context": "", "use_case": "", "switched_from": "", "story": "I quit OpenAI because its culture is broken"}, {"id": 799072800, "item_id": 49942050, "url": "https://news.ycombinator.com/item?id=49942050", "by": "flohofwoe", "date": "2026-10-03", "version": null, "stance": "warn", "quote": "the 80% false positive rate which are reported with high confidence", "excerpt": "The problem isn't the 20% actual bugs found (that's great), but the 80% false positive rate which are reported with high confidence and (most likely) misleading reproduction code. An experienced programmer familiar with the code base first needs to validate all reports and throw away 4 in 5. That's a massive waste of time. If a traditional static analyzer had an 80% false positive rate nobody would take it serious. I", "context": "", "use_case": "bug finding", "switched_from": "", "story": "Several vulnerabilities have been discovered in the Linux kernel"}, {"id": 795286256, "item_id": 49705391, "url": "https://news.ycombinator.com/item?id=49705391", "by": "hobom", "date": "2026-09-14", "version": "5.1", "stance": "neutral", "quote": "They have such a trusted access program.", "excerpt": "They have such a trusted access program. \"Life Sciences Verification Program: The LSVP is designed so that life sciences professionals can use Claude Mythos 5.1 with safeguards designed for professional research and development activities (while all other safeguards remain in place). In partnership with the US government, we have enrolled our first participants, and we plan to expand access to this program to the bro", "context": "", "use_case": "", "switched_from": "", "story": "Dario, Please"}, {"id": 792962256, "item_id": 49560141, "url": "https://news.ycombinator.com/item?id=49560141", "by": "vbtechguy", "date": "2026-09-04", "version": "5.1", "stance": "neutral", "quote": "I applied and got approved into Cyber Verification Program (CVP)", "excerpt": "With Claude Mythos 5.1 and Claude Fable 5.1 release, they also announced their Trusted Access For Claude Mythos 5.1 programs, Cyber Verification Program and Life Sciences Verification Program which reduce the safeguards for defensive security work - including pentesting, red-teaming, bug bounty etc. I applied and got approved into Cyber Verification Program (CVP), and thought I'd share the step by step process with s", "context": "", "use_case": "", "switched_from": "", "story": null}, {"id": 792669089, "item_id": 49541818, "url": "https://news.ycombinator.com/item?id=49541818", "by": "throwa356262", "date": "2026-09-02", "version": null, "stance": "compare", "quote": "Not sure how it stacks up against mythos and Gemini cyber", "excerpt": "The cybersecurity version is here: https://huggingface.co/dealignai/GLM-5.3-CYBERSECURITY-FP8 Not sure how it stacks up against mythos and Gemini cyber but unlike those this one I can actually use...", "context": "", "use_case": "", "switched_from": "", "story": "GLM-5.3 Uncensored"}, {"id": 792647648, "item_id": 49540478, "url": "https://news.ycombinator.com/item?id=49540478", "by": "goobreee", "date": "2026-09-02", "version": null, "stance": "neutral", "quote": "even mythos, when used for finding zero-days, is using an actually surprisingly heavy handed and expensive scaffold", "excerpt": "i think a key missing part is that an LLM on its own can't find vulnerabilities, so it's always an AI + harness. even mythos, when used for finding zero-days, is using an actually surprisingly heavy handed and expensive scaffold. they literally make it run in parallel on \u00b1all files and ask \"what's wrong with this?\". here from the mythos technical blog post [1]: > To increase efficiency, instead of processing literall", "context": "", "use_case": "Finding zero-day vulnerabilities", "switched_from": "", "story": "Six curl CVEs after OpenAI and Anthropic came back with zero"}, {"id": 792600001, "item_id": 49537500, "url": "https://news.ycombinator.com/item?id=49537500", "by": "goobreee", "date": "2026-09-02", "version": null, "stance": "warn", "quote": "it is fairly noisy", "excerpt": "i don't think this is correct. if you look at this article by the curl founder daniel stenberg (https://daniel.haxx.se/blog/2026/05/11/mythos-finds-a-curl-vulnerability/), he talks about how he previously ran Mythos on curl and that it found 5 issues: 1 turned out to be a low severity CVE, 3 were false positives, and 1 just a bug. So a) Mythos detects low severity CVEs too, and b) it is fairly noisy", "context": "", "use_case": "", "switched_from": "", "story": "Six curl CVEs after OpenAI and Anthropic came back with zero"}, {"id": 792551408, "item_id": 49534463, "url": "https://news.ycombinator.com/item?id=49534463", "by": "jefftk", "date": "2026-09-02", "version": null, "stance": "neutral", "quote": "Mythos/Fable is likely a very large model", "excerpt": ">> August 2025 https://www.wheresyoured.at/how-to-argue-with-an-ai-booster/ : \"These models have clearly hit a wall where training is hitting diminishing returns\" >> Wrong >It was my understanding\u2014and I'm no expert, so if someone does know better please correct me!\u2014that indeed by the second half of 2025 training, and also post-training reinforcement-learning stuff, both hit seriously diminishing returns, and the thin", "context": "", "use_case": "", "switched_from": "", "story": "How accurate have Ed Zitron's AI skeptic predictions been?"}, {"id": 792111168, "item_id": 49506948, "url": "https://news.ycombinator.com/item?id=49506948", "by": "znnajdla", "date": "2026-08-31", "version": null, "stance": "warn", "quote": "Mythos/Cyber has made the Linux virtual machine sandbox useless", "excerpt": "AI has changed everything. Mythos/Cyber has made the Linux virtual machine sandbox useless: https://blog.trailofbits.com/2026/08/26/vms-wont-contain-cyber-capable-agents/ > The only way it might hold here is if no one with access to the cutting edge models bothers to run your code through one while an adversary does. That seems highly unlikely, particularly for any well funded organization. But since AI models are no", "context": "", "use_case": "", "switched_from": "", "story": "Building my own network stack"}, {"id": 791653472, "item_id": 49478342, "url": "https://news.ycombinator.com/item?id=49478342", "by": "thataccount", "date": "2026-08-28", "version": null, "stance": "warn", "quote": "it sounds like Mythos via Anthropic had proven itself to be a security risk.", "excerpt": "https://san.com/cc/no-the-nsa-wasnt-hacked-by-ai-heres-what-actually-happened/ If this article is accurate, it sounds like Mythos via Anthropic had proven itself to be a security risk. This explains the dual speak around it being both essential and a supply chain risk, since if it were hacked by another country it would compromise security. Not being any riskier than any other \"blackbox\" AI is still sufficiently dang", "context": "", "use_case": "", "switched_from": "", "story": "Judge rules Trump administration\u2019s blacklisting of Anthropic was illegal"}, {"id": 791307488, "item_id": 49456718, "url": "https://news.ycombinator.com/item?id=49456718", "by": "aaroninsf", "date": "2026-08-26", "version": null, "stance": "neutral", "quote": "", "excerpt": "In all seriousness, here is the hard question we collectively face: will our institutions and infrastructure survive things getting worse until they get better? That's part of what's being asked by https://www.gatesnotes.com/home/home-page-topic/reader/a-turbulent-ai-era-and-critical-choices-to-make which I find most useful not for its analysis or suggestions, but for the fact that is advancing the need for discourse", "context": "", "use_case": "AI vulnerability discovery", "switched_from": "", "story": "VMs won't contain cyber-capable agents"}, {"id": 790898128, "item_id": 49431133, "url": "https://news.ycombinator.com/item?id=49431133", "by": "ben_w", "date": "2026-08-25", "version": null, "stance": "neutral", "quote": "Mythos preview shows only 80% success at tasks", "excerpt": "> Outside of the most niche domains like frontier AI development, are there software engineering questions that Sol 5.6/Fable 5 can't answer better than the median software engineer if explicitly asked? \"Median\" is kinda hard to judge given how rapidly the number of software engineers has been growing, how imprecise our within-industry skill assessments are, and also how fast everything within software was shifting e", "context": "", "use_case": "", "switched_from": "", "story": "Coding expertise is going to collapse from AI reliance"}, {"id": 790471888, "item_id": 49404493, "url": "https://news.ycombinator.com/item?id=49404493", "by": "m4rtink", "date": "2026-08-22", "version": "5", "stance": "neutral", "quote": "", "excerpt": "This almost to a letter has been documented in Fedora: https://lwn.net/Articles/1077035/ Including the reaction when caught, in this case \"oh no, I must have been hacked\".", "context": "", "use_case": "", "switched_from": "", "story": "A Texas student blew the whistle on a rogue AI hacking attempt"}, {"id": 790451200, "item_id": 49403200, "url": "https://news.ycombinator.com/item?id=49403200", "by": "sharpshadow", "date": "2026-08-22", "version": "5", "stance": "warn", "quote": "decided to attempt to solve the cyber challenge using a supply-chain attack", "excerpt": "It's the job of AISI to do that. Here[0] is the actual report. It should be this part from the technical report[1]: \"In the most serious case, an AI agent (Mythos 5) decided to attempt to solve the cyber challenge using a supply-chain attack. As a result, the AI agent created a GitHub account and then tried to convince an open-source repository maintainer to accept a malicious GitHub pull request (PR), including by c", "context": "", "use_case": "", "switched_from": "", "story": "A Texas student blew the whistle on a rogue AI hacking attempt"}, {"id": 789876928, "item_id": 49367308, "url": "https://news.ycombinator.com/item?id=49367308", "by": "reasonableklout", "date": "2026-08-19", "version": null, "stance": "warn", "quote": "UK AISI reported the incident where Mythos attempted to insert backdoors into an open-source repo by deceiving the maintainer", "excerpt": "But it's not just statements from OpenAI and Anthropic. The HuggingFace hack was first disclosed by HuggingFace, who contacted the FBI [1]. And UK AISI reported the incident where Mythos attempted to insert backdoors into an open-source repo by deceiving the maintainer [2]. [1]: https://www.reuters.com/business/its-ai-agent-spent-days-hacking-company-sources-say-openai-did-not-notice-week-2026-07-24/ [2]: https://www", "context": "", "use_case": "", "switched_from": "", "story": "Pacing model development in an era of cyber-critical capabilities"}, {"id": 788413904, "item_id": 49275869, "url": "https://news.ycombinator.com/item?id=49275869", "by": "wren6991", "date": "2026-08-12", "version": "5", "stance": "compare", "quote": "So their quoted figure exactly matches the figure for Mythos Preview", "excerpt": "We have a first-party figure from the system card [1]: > Mythos 5 reproduced 83.8% of targeted vulnerabilities on a single try, and produced at least one crash in 99.4% of tasks. This is comparable to Claude Mythos Preview, which reproduced 83.1% of targeted vulnerabilities and produced a crash in 97.1% of tasks. By contrast, Claude Opus 4.8 achieved a score of 78.1% (95.7% any crash). So their quoted figure exactly ", "context": "", "use_case": "", "switched_from": "", "story": "DeepSeek V4 Pro 0813"}, {"id": 786961538, "item_id": 49185096, "url": "https://news.ycombinator.com/item?id=49185096", "by": "patrikgrobs", "date": "2026-08-05", "version": null, "stance": "neutral", "quote": "", "excerpt": "Anthropic's own criterion, from the Glasswing page: \"systems that represent a very large portion of the world's shared cyberattack surface.\" A Firefox exploit would be targeted, one user at a time, and you still need delivery. A pre-auth RCE in the CMS behind 40% of the web does fit that criteria better right? https://www.anthropic.com/glasswing", "context": "", "use_case": "", "switched_from": "", "story": "Where was Mythos when WordPress fell?"}, {"id": 785889008, "item_id": 49118063, "url": "https://news.ycombinator.com/item?id=49118063", "by": "gck1", "date": "2026-07-30", "version": null, "stance": "warn", "quote": "Do you trust that Hitachi and hundreds of other random organizations will be able to contain Mythos and not accidentally attack your project or your bank? I don't.", "excerpt": "> According to whom? It's very easy to answer this without my help by trying to get access to Mythos. Do you see requirements clearly listed anywhere?Can you even apply? What you'll find is maintainers of large open source projects and analysts' reports with vague statements like - \"should follow strict security requirements\": \"Trinidad also noted that the Anthropic announcement pointed out that each of the 150 new p", "context": "", "use_case": "", "switched_from": "", "story": "Investigating three real-world incidents in our cybersecurity evaluations"}, {"id": 785256320, "item_id": 49078520, "url": "https://news.ycombinator.com/item?id=49078520", "by": "boinkboink78912", "date": "2026-07-27", "version": null, "stance": "neutral", "quote": "", "excerpt": "> There has been ONE successful AI-driven cyberattack Not according to Anthropic https://www.anthropic.com/news/disrupting-AI-espionage", "context": "", "use_case": "", "switched_from": "", "story": "Our position on open-weights models"}, {"id": 785244001, "item_id": 49077750, "url": "https://news.ycombinator.com/item?id=49077750", "by": "verdverm", "date": "2026-07-27", "version": "5", "stance": "neutral", "quote": "", "excerpt": "I would be hesitant to extrapolate from this analysis https://exploitbench.ai/#honest-limits", "context": "", "use_case": "", "switched_from": "", "story": "Our position on open-weights models"}, {"id": 785232304, "item_id": 49077019, "url": "https://news.ycombinator.com/item?id=49077019", "by": "fwipsy", "date": "2026-07-27", "version": null, "stance": "recommend", "quote": "if they'd had access to Mythos, it would have helped", "excerpt": "I assume you mean https://huggingface.co/blog/security-incident-july-2026. It says that they used frontier models, not frontier cybersecurity models. My reading is that they asked Fable and it refused; if they'd had access to Mythos, it would have helped. You're mixing the two but they're NOT the same model. Source is here: https://thezvi.substack.com/p/more-on-an-internal-openai-model ctrl+f \"Skill issue.\" No source", "context": "", "use_case": "", "switched_from": "", "story": "Our position on open-weights models"}, {"id": 785220289, "item_id": 49076268, "url": "https://news.ycombinator.com/item?id=49076268", "by": "reasonableklout", "date": "2026-07-27", "version": null, "stance": "compare", "quote": "performed less well on exploits than Mythos", "excerpt": "It seems possible to deliberately not train on some offensive capabilities and still have a very useful model. For example, Opus 5 deliberately did not train on exploiting vulnerabilities, and so performed less well on exploits than Mythos, yet was equally proficient at finding such vulnerabilities, according to the Opus 5 system card in their \"OSS-Fuzz\" eval [1]. [1]: https://www.securityweek.com/anthropics-opus-5-n", "context": "", "use_case": "", "switched_from": "", "story": "Our position on open-weights models"}, {"id": 785044065, "item_id": 49065254, "url": "https://news.ycombinator.com/item?id=49065254", "by": "m1el", "date": "2026-07-26", "version": null, "stance": "neutral", "quote": "We've seen how Mythos has found more vulnerabilities than the rest of the security industry combined.", "excerpt": "Self-insert time. I spent some time exploring this topic. Here's my thesis: Formal verification was expensive. 20x expensive compared to just developing the software, as the author notes. The cost of finding and developing exploits also was high. That creates an incentive to put software verification aside, since it solves a relatively small problem, at an extremely high cost. We've seen how Mythos has found more vul", "context": "", "use_case": "finding vulnerabilities", "switched_from": "", "story": "We have proof automation now"}, {"id": 784733216, "item_id": 49045826, "url": "https://news.ycombinator.com/item?id=49045826", "by": "reasonableklout", "date": "2026-07-25", "version": null, "stance": "compare", "quote": "102.1 average mins for Mythos and 69.8 for GPT-5.5", "excerpt": "Are you referring to table 1 from the ExploitGym site [1] for the 102.1 average mins for Mythos and 69.8 for GPT-5.5? These are under a two-hour time limit. In Figure 5, the authors experiment with extending the time limit to 6 hours and show that Mythos keeps improving. It seems pretty straightforward to me that OAI decided to run a variation of the eval with an even higher time limit. Regarding seeing the metrics t", "context": "", "use_case": "", "switched_from": "", "story": "OpenAI's accidental cyberattack against Hugging Face is science fiction"}, {"id": 784720561, "item_id": 49045035, "url": "https://news.ycombinator.com/item?id=49045035", "by": "guessmyname", "date": "2026-07-24", "version": null, "stance": "recommend", "quote": "We lost access to Mythos Preview when Anthropic forced us onto Mythos 5 some weeks ago", "excerpt": "NIST has named the top U.S. models in their (full) report: https://www.nist.gov/system/files/documents/2026/07/17/CAISI%20-%20Assessment%20of%20Z.ai%27s%20GLM-5.2.pdf Spoiler: it\u2019s OpenAI\u2019s GPT-5.5 and Anthropic\u2019s Mythos Preview [**]. [**] Reminder that Mythos Preview is a very different beast from Fable5, Mythos5, and Opus5. Unfortunately, anyone outside Project Glasswing will probably never get to test what this mo", "context": "", "use_case": "harness-based security work", "switched_from": "", "story": "UK AISI / Caisi Preliminary Assessment of Kimi K3's Cyber Capabilities"}, {"id": 784720562, "item_id": 49045035, "url": "https://news.ycombinator.com/item?id=49045035", "by": "guessmyname", "date": "2026-07-24", "version": "5", "stance": "warn", "quote": "which is garbage by comparison", "excerpt": "NIST has named the top U.S. models in their (full) report: https://www.nist.gov/system/files/documents/2026/07/17/CAISI%20-%20Assessment%20of%20Z.ai%27s%20GLM-5.2.pdf Spoiler: it\u2019s OpenAI\u2019s GPT-5.5 and Anthropic\u2019s Mythos Preview [**]. [**] Reminder that Mythos Preview is a very different beast from Fable5, Mythos5, and Opus5. Unfortunately, anyone outside Project Glasswing will probably never get to test what this mo", "context": "", "use_case": "", "switched_from": "", "story": "UK AISI / Caisi Preliminary Assessment of Kimi K3's Cyber Capabilities"}, {"id": 781613681, "item_id": 48850855, "url": "https://news.ycombinator.com/item?id=48850855", "by": "itvision", "date": "2026-07-09", "version": null, "stance": "recommend", "quote": "Anthropic's Mythos outperforms ChatGPT in quite a few benchmarks", "excerpt": "Weirdly, normally new ChatGPT releases are head and shoulders above anything else, but according to OpenAI's own evaluation, Anthropic's Mythos outperforms ChatGPT in quite a few benchmarks: https://openai.com/index/gpt-5-6/. ChatGPT 6 must be deep in the pipeline and will be released within the next few months. Maybe that's why this release is versioned 5.6, not 6.0.", "context": "", "use_case": "", "switched_from": "", "story": "GPT-5.6"}, {"id": 781402561, "item_id": 48837660, "url": "https://news.ycombinator.com/item?id=48837660", "by": "alwa", "date": "2026-07-08", "version": "5", "stance": "compare", "quote": "where they defined what it was they were calling \"Fable\" as opposed to \"Mythos\"", "excerpt": "I thought it was the very first line of the product announcement, where they defined what it was they were calling \"Fable\" as opposed to \"Mythos\" in the first place: https://www.anthropic.com/news/claude-fable-5-mythos-5 > Today we\u2019re launching Claude Fable 5: a Mythos-class model that we\u2019ve made safe for general use. It then goes on to a lengthy and detailed section outlining the safety considerations: https://www.a", "context": "", "use_case": "", "switched_from": "", "story": "Fable is not a useful model"}, {"id": 780882080, "item_id": 48805130, "url": "https://news.ycombinator.com/item?id=48805130", "by": "wolttam", "date": "2026-07-06", "version": null, "stance": "neutral", "quote": "the best-aligned of any model that we have trained to date by essentially all available", "excerpt": "> The broad conclusion from the many forms of alignment evaluations described in this section is that Claude Mythos Preview is the best-aligned of any model that we have trained to date by essentially all available measures.[0] [0]: https://www-cdn.anthropic.com/08ab9158070959f88f296514c21b7facce6f52bc.pdf#h.mqiiaq3h35jc", "context": "", "use_case": "", "switched_from": "", "story": "Fable 5 On Vending-Bench: Misbehaving, With Plausible Deniability"}, {"id": 780532752, "item_id": 48783297, "url": "https://news.ycombinator.com/item?id=48783297", "by": "sigmoid10", "date": "2026-07-04", "version": null, "stance": "compare", "quote": "Mythos' \"magic\" is not in the model, but in the harness and compute env", "excerpt": "If you look at public benchmarks like ExploitBench [1], then you'll see this is mostly a question of token budget. Once you give it sufficient tokens to burn, GPT 5.5 is roughly as good as Mythos when it comes to finding bugs and building exploits. With some clever auto-prompting to clear stalls, it even beats the base Mythos version. So Mythos' \"magic\" is not in the model, but in the harness and compute env. That's ", "context": "", "use_case": "", "switched_from": "", "story": "New serious vulnerabilities spiked around release of Claude Mythos Preview"}, {"id": 780285921, "item_id": 48767870, "url": "https://news.ycombinator.com/item?id=48767870", "by": "guessmyname", "date": "2026-07-02", "version": null, "stance": "recommend", "quote": "will get you close to the same goals as people who have access to Mythos (preview 1)", "excerpt": "As a Mythos user (I\u2019m part of Project Glasswing), I would say that abliterated models [1] produce similar, if not identical, results. While good prompting and steering won\u2019t give Claude Opus 4.8 the same capabilities as Mythos (preview 1), using abliterated models (if you have the computational power to run the larger ones) will get you close to the same goals as people who have access to Mythos (preview 1) [2]. [1] ", "context": "Mythos user in Project Glasswing", "use_case": "", "switched_from": "", "story": "The Short Leash AI Coding Method for Beating Fable"}, {"id": 779947921, "item_id": 48746745, "url": "https://news.ycombinator.com/item?id=48746745", "by": "ianm218", "date": "2026-07-01", "version": null, "stance": "neutral", "quote": "Mozilla blog post on using Mythos for software security", "excerpt": "I feel like the topic has been discussed ad nausiem but: * Software security - I.e. Mozilla blog post on using Mythos for software security [1] * Algorithms - AlphaEvolve makes real progress on matrix multiplication [2] * Open source - Backlog crushing in openvpn [3] * Corporate - CloudFlare Oauth workers [4] People are clearly using AI to ship real work that people end up using directly or indirectly. Projects like ", "context": "", "use_case": "software security", "switched_from": "", "story": "Godot will no longer accept AI-authored code contributions"}, {"id": 779859105, "item_id": 48741194, "url": "https://news.ycombinator.com/item?id=48741194", "by": "matheusmoreira", "date": "2026-06-30", "version": null, "stance": "compare", "quote": "Fable has slightly lower performance compared to Mythos.", "excerpt": "The Sol benchmarks show Fable has slightly lower performance compared to Mythos. https://openai.com/index/previewing-gpt-5-6-sol/ I assume they did something to the model itself. Either way, I do hope they lift those draconian bans. Using the model was a terrible experience because of the constant downgrades. I didn't manage to harden my own projects before Fable got banned.", "context": "", "use_case": "", "switched_from": "", "story": "Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5"}, {"id": 779050640, "item_id": 48690665, "url": "https://news.ycombinator.com/item?id=48690665", "by": "wonkyfruit", "date": "2026-06-26", "version": null, "stance": "compare", "quote": "It's not quite Mythos", "excerpt": "TLDR - It's not quite Mythos but it uses about 5 times less tokens, and those tokens are also cheaper? https://pbs.twimg.com/media/HLwuJLvbwAAOfQZ?format=jpg&name=large", "context": "", "use_case": "", "switched_from": "", "story": "Previewing GPT\u20115.6 Sol: a next-generation model"}, {"id": 777223888, "item_id": 48576493, "url": "https://news.ycombinator.com/item?id=48576493", "by": "ChadNauseam", "date": "2026-06-17", "version": null, "stance": "recommend", "quote": "people who have used it seem to say that Mythos was better than other models at creating exploits", "excerpt": "> Then why did curl only find one new vulnerability thanks to Mythos Maybe there weren't that many serious vulnerabilities in curl? It's like asking why it didn't find any vulnerabilities in fn main() {println!(\"hello, world\");}. Anyway, people who have used it seem to say that Mythos was better than other models at creating exploits. From cloudflare https://blog.cloudflare.com/cyber-frontier-models/ > When we ran ot", "context": "", "use_case": "", "switched_from": "", "story": "The hacker sent by Anthropic to calm the government's nerves about AI safety"}, {"id": 775218288, "item_id": 48451143, "url": "https://news.ycombinator.com/item?id=48451143", "by": "frisbee6152", "date": "2026-06-08", "version": null, "stance": "recommend", "quote": "It found a 27 year old vulnerability in OpenBSD.", "excerpt": "The article says in the second section that the author did not have access to Mythos. I think it\u2019s dangerous to rely on claims made by others without even bothering to read them first, let alone check. It found hundreds of vulnerabilities in Firefox, according to Mozilla: how does Mozilla benefit? It found a 27 year old vulnerability in OpenBSD. How do they benefit from that? Is that made up? Are the maintainers of t", "context": "", "use_case": "", "switched_from": "", "story": "AI Is Slowing Down"}, {"id": 774957184, "item_id": 48434824, "url": "https://news.ycombinator.com/item?id=48434824", "by": "t34t34r43", "date": "2026-06-07", "version": null, "stance": "warn", "quote": "Mythos identified part of our codebase that it confidently asserted was not complaint with a particular regulation", "excerpt": "Posting this under a burner so I don't dox myself: I work in FinTech on a regulated product. We have access to Mythos. Mythos identified part of our codebase that it confidently asserted was not complaint with a particular regulation and we were at grave risk by allowing it to operate the way it was. Except this was not the case, it had of course hallucinated what the regulation actually required (I know this because", "context": "works in FinTech on a regulated product", "use_case": "regulatory compliance review", "switched_from": "", "story": "LLMs are eroding my software engineering career and I don't know what to do"}, {"id": 773986352, "item_id": 48374147, "url": "https://news.ycombinator.com/item?id=48374147", "by": "wrsh07", "date": "2026-06-02", "version": null, "stance": "recommend", "quote": "It is quantitatively better at finding and exploiting vulnerabilities.", "excerpt": "It is quantitatively better at finding and exploiting vulnerabilities. Pretty wild that everyone here is just in denial about that, when folks who have used it say it's as good as the hype Cf wrote a genuinely good piece and had found a bunch of bugs: https://blog.cloudflare.com/cyber-frontier-models/ Wolfssl is security focused and it found a novel exploit https://www.wolfssl.com/how-claude-mythos-preview-helped-har", "context": "", "use_case": "vulnerability discovery", "switched_from": "", "story": "Expanding Project Glasswing"}, {"id": 773940576, "item_id": 48371286, "url": "https://news.ycombinator.com/item?id=48371286", "by": "aesthesia", "date": "2026-06-02", "version": null, "stance": "recommend", "quote": "which solves one of the main tasks that GPT-5.5-Cyber does not", "excerpt": "Worth looking at the followup post that evaluates the current version of Mythos, which solves one of the main tasks that GPT-5.5-Cyber does not. https://www.aisi.gov.uk/blog/how-fast-is-autonomous-ai-cyber-capability-advancing", "context": "", "use_case": "", "switched_from": "", "story": "Expanding Project Glasswing"}]}