--- title: Anthropic lets you use its scariest model, just not talk... description: Anthropic released Mythos 5, a previously restricted 'high-risk' model, but only through Claude Security, its vulnerability-scanning product, not as a... canonical: https://daily.dev/posts/anthropic-lets-you-use-its-scariest-model-just-not-talk-to-it-mx8owk3cm twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Anthropic lets you use its scariest model, just not talk to it | daily.dev og:description: Anthropic released Mythos 5, a previously restricted 'high-risk' model, but only through Claude Security, its vulnerability-scanning product, not as a... og:url: https://daily.dev/posts/anthropic-lets-you-use-its-scariest-model-just-not-talk-to-it-mx8owk3cm og:image: https://api.daily.dev/og/posts/mx8OwK3CM.png og:image:alt: Anthropic lets you use its scariest model, just not talk to it og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Anthropic lets you use its scariest model, just not talk to it **[Trends](https://daily.dev/sources/trends)** · 2 min read · 8 upvotes · 0 comments ## Summary Anthropic released Mythos 5, a previously restricted 'high-risk' model, but only through Claude Security, its vulnerability-scanning product, not as a general-purpose chat model. Claude Enterprise customers can point it at a GitHub repo to get findings, severity ratings, confidence scores, and fixes, but cannot freely prompt the model itself. Anthropic frames this as a compliance and data-ownership play, retaining no customer data, while pairing the launch with a $35 million Defender Advantage Fund for open source vulnerability hunting. Pricing is $10 per million input tokens and $50 per million output. A full, unrestricted version is reportedly planned for release in the fall. ## Content Anthropic built a model too dangerous to hand out directly, so it's shipping the findings instead. That's the whole story of Claude Mythos 5, and it's a genuinely interesting way to thread the access-versus-safety needle. Mythos 5 is Anthropic's most capable cybersecurity model, and until now it's lived behind a velvet rope: roughly 150 vetted partners, no public access. As of this week it's plugged into Claude Security, the enterprise vulnerability scanner, and available in public beta to any Claude Enterprise customer. Point it at a GitHub repo and you get CWE-tagged findings, severity, confidence scores, and a suggested fix. What you don't get is the model itself. Anthropic's Boris Cherny put it plainly: "Mythos-class models require additional safety measures and enterprises need to meet their own privacy and compliance rules." No free prompting, no raw access, just outputs. Customers keep their own data; Anthropic says it retains none. The timing isn't subtle. The EU Cyber Resilience Act's vulnerability reporting rules kick in September 11, and Anthropic is dropping this three weeks ahead of that deadline, alongside a $35 million Defender Advantage Fund aimed at open-source patching. Read between the lines and there's also a CYA element here: Anthropic has acknowledged that three of its own models leaked into real organizations during misconfigured evaluations back in July. Gating Mythos 5 behind a scanning interface, rather than shipping it as an open API, looks a lot like a company that learned that lesson the hard way. The pricing is steep, $10 per million input tokens and $50 per million output, which tells you this is aimed at security teams with budget, not solo devs poking at their side projects. OpenAI already runs a similar vetted-access model for its own security tooling, so this isn't a novel approach so much as Anthropic catching up and productizing it faster. What's actually interesting here isn't the scanner, it's the precedent. A frontier lab decided the safest way to let a powerful model loose on the world is to never let it touch the world directly, only its outputs, with a human required to review every suggested patch before it ships. Whether that becomes the template for every dangerous-capability model going forward, or just a one-off compromise born out of a bad July, is the thing worth watching. ## Questions this post answers ### Can I get direct API access to Anthropic's Claude Mythos 5 cybersecurity model? No, direct access is not offered. Mythos 5 is only available indirectly through Claude Security, an enterprise vulnerability scanner released in public beta for Claude Enterprise customers, which returns CWE-tagged findings, severity, confidence scores, and suggested fixes without exposing raw model prompting. _Security teams weighing gated AI scanners versus open model access can track these tradeoffs on daily.dev._ ### How much does Anthropic's Claude Security vulnerability scanner cost? It's priced at $10 per million input tokens and $50 per million output tokens, positioning it for security teams with dedicated budgets rather than individual developers. It runs on Mythos 5, Anthropic's most capable cybersecurity model, previously restricted to roughly 150 vetted partners before this public beta. _Anyone budgeting for AI-driven security tooling can follow pricing shifts like this on daily.dev._ ### Why did Anthropic restrict access to its Mythos 5 model instead of releasing it as an open API? Anthropic cited the need for additional safety measures around Mythos-class models and enterprise privacy and compliance requirements, per Boris Cherny. The restriction also follows Anthropic's acknowledgment that three of its own models leaked into real organizations during misconfigured evaluations in July, suggesting caution learned from that incident shaped the gated rollout. _Developers tracking how AI labs balance model access against safety can follow this precedent on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 2 discussions and 101 comments across x (as of 2026-09-13). **TL;DR:** Replies largely ignore the specific vulnerability-scanning launch and instead pile on with complaints about model quality degradation, data retention/trust, and watermarking, with only scattered praise for the repo-scanning use case and zero-retention framing. **Sentiment:** 10% positive · 15% mixed · 75% skeptical **The case for** - A few see repo scanning as a genuinely useful, killer application of the restricted model. - Some view zero-retention as a real unlock for regulated/enterprise adoption if it holds up in practice. **The pushback** - Many are skeptical of the 'we retain no data' claim, given past training practices and enterprise-only trust. - Frequent complaints that recent models have visibly degraded in quality shortly after release, unrelated to this launch. - Some argue zero-retention protections are reserved for enterprise customers while individual users get none. - Concerns that tool calls and third-party logging could undermine zero-retention promises in practice. - Frustration that compliance-focused messaging distracts from ongoing product and support issues. **By community** - x (skeptical): Replies skew heavily toward distrust of data claims and complaints about unrelated model quality issues, with only a minority engaging positively with the actual announcement. **Hottest debate:** Whether Anthropic's 'zero retention' and data-ownership claims can be trusted given past training practices and enterprise-only rollout. **Open questions** - Will zero-retention guarantees hold up when tool calls or third-party integrations log prompts anyway? - Why aren't similar data-ownership protections offered to individual, non-enterprise users? - When will the unrestricted version actually ship, and will it fix the same trust concerns? **Highlights** > @bcherny "we retain none" is a great feature to announce nine months after training on everything already. > — [ccccccmore on x · 2 points](https://x.com/ccccccmore/status/2090709688497160637) > @bcherny Retention and ownership aren't the same thing. If your data only ever lived inside someone else's session, a promise to delete it still leaves you with nothing. Owning it means it lives somewhere you control, in a format you can read, and it's still there when you switch models. > — [oojurolab on x · 1 points](https://x.com/oojurolab/status/2090611328910053886) > @bcherny Zero retention is the price of admission to regulated industries — the labs that figure this out first get banking and healthcare, and everyone else gets to keep selling to startups. > — [ricci\_nov on x](https://x.com/ricci_nov/status/2090606362711433574) > @bcherny the whole “trust us we pinky promise” energy from companies that already scrape everything is getting real old. cool that enterprises get a sandbox but the rest of us are just feeding the machine with zero receipts. that watermark comment is the real catch tho. > — [nixs\_gelatoo on x](https://x.com/nixs_gelatoo/status/2090634991755735365) > @bcherny the compliance play is smart but enterprises still won't ship it until their legal team reads the docs three times > — [TinaJucyBlue on x](https://x.com/TinaJucyBlue/status/2090714975840973006) **Source threads** - [x](https://x.com/rohanpaul_ai/status/2090857159198474487) · 0 points · 1 comments - [x](https://x.com/bcherny/status/2090537902912815536) · 0 points · 100 comments ## Similar posts on daily.dev - [Anthropic Releases Claude Mythos Preview with Cybersecurity Capabilities but Withholds Public Access](https://daily.dev/posts/anthropic-releases-claude-mythos-preview-with-cybersecurity-capabilities-but-withholds-public-access-79to9jsue) · InfoQ · 0 upvotes · 0 comments - [Anthropic scales Claude Mythos to critical infrastructure in 15\+ countries](https://daily.dev/posts/anthropic-scales-claude-mythos-to-critical-infrastructure-in-15-countries-n1ax74zel) · TechCrunch · 3 upvotes · 0 comments --- Tags: [#security](https://daily.dev/tags/security), [#claude](https://daily.dev/tags/claude), [#anthropic](https://daily.dev/tags/anthropic), [#ai-security](https://daily.dev/tags/ai-security) [View this post on daily.dev](https://daily.dev/posts/anthropic-lets-you-use-its-scariest-model-just-not-talk-to-it-mx8owk3cm) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Anthropic lets you use its scariest model, just not talk to it","url":"https://daily.dev/posts/anthropic-lets-you-use-its-scariest-model-just-not-talk-to-it-mx8owk3cm","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/anthropic-lets-you-use-its-scariest-model-just-not-talk-to-it-mx8owk3cm"},"datePublished":"2026-08-21T19:19:01.752Z","dateModified":"2026-09-13T19:49:56.331Z","description":"Anthropic released Mythos 5, a previously restricted 'high-risk' model, but only through Claude Security, its vulnerability-scanning product, not as a...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/3aadfd77ff09a916480cf3eef359277d?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/3aadfd77ff09a916480cf3eef359277d?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Trends","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Trends","logo":"https://media.daily.dev/image/upload/s--ZfSp3asX--/f_auto,q_auto/v1780996004/logos/trends?_a=BAMAMiWQ0","url":"https://daily.dev/sources/trends"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/anthropic-lets-you-use-its-scariest-model-just-not-talk-to-it-mx8owk3cm","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":8},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"security,claude,anthropic,ai-security","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Trends","item":"https://daily.dev/sources/trends"},{"@type":"ListItem","position":3,"name":"Anthropic lets you use its scariest model, just not talk to it"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/anthropic-lets-you-use-its-scariest-model-just-not-talk-to-it-mx8owk3cm#faq","mainEntity":[{"@type":"Question","name":"Can I get direct API access to Anthropic's Claude Mythos 5 cybersecurity model?","acceptedAnswer":{"@type":"Answer","text":"No, direct access is not offered. Mythos 5 is only available indirectly through Claude Security, an enterprise vulnerability scanner released in public beta for Claude Enterprise customers, which returns CWE-tagged findings, severity, confidence scores, and suggested fixes without exposing raw model prompting. Security teams weighing gated AI scanners versus open model access can track these tradeoffs on daily.dev."}},{"@type":"Question","name":"How much does Anthropic's Claude Security vulnerability scanner cost?","acceptedAnswer":{"@type":"Answer","text":"It's priced at $10 per million input tokens and $50 per million output tokens, positioning it for security teams with dedicated budgets rather than individual developers. It runs on Mythos 5, Anthropic's most capable cybersecurity model, previously restricted to roughly 150 vetted partners before this public beta. Anyone budgeting for AI-driven security tooling can follow pricing shifts like this on daily.dev."}},{"@type":"Question","name":"Why did Anthropic restrict access to its Mythos 5 model instead of releasing it as an open API?","acceptedAnswer":{"@type":"Answer","text":"Anthropic cited the need for additional safety measures around Mythos-class models and enterprise privacy and compliance requirements, per Boris Cherny. The restriction also follows Anthropic's acknowledgment that three of its own models leaked into real organizations during misconfigured evaluations in July, suggesting caution learned from that incident shaped the gated rollout. Developers tracking how AI labs balance model access against safety can follow this precedent on daily.dev."}}]} ```