{"rewrite":{"id":"r_5fe2b41d94552077d9ade23c","clusterId":"c_64bcd358f38f12a5ed80144b","slug":"claude-mythos-5-widens-reach-via-claude-security-and-external-services","model":"deepseek-v4-flash","headline":"Claude Mythos 5 Widens Reach via Claude Security and External Services","summary":"Anthropic has expanded availability of its cybersecurity-focused AI model Claude Mythos 5, which was previously restricted to select organizations due to misuse concerns. The model is now integrated into the security product Claude Security, a vulnerability analysis service in public beta for Claude Enterprise users. To reduce abuse risk, Mythos 5 operates only for predetermined behind-the-scenes processing, with end users receiving only fixed outputs such as vulnerability fix suggestions and security alerts. Specifying a GitHub repository prompts Mythos 5 to examine source code and report findings organized by CWE type, confidence, severity, and recommended fixes. Fix suggestions can be implemented in the web version of Claude, though interactive fix work uses models available to the organization rather than Mythos 5. Alongside this expansion, Anthropic announced the Defender Advantage Fund, which will provide Claude usage credits totaling $35 million to support fixing vulnerabilities in widely used open-source software. The Cyber Verification Program for security organizations that pass review will relax restrictions on cybersecurity-related capabilities in the model and the model, with defensive capabilities using Mythos-class models planned for the future. Anthropic states it will develop safety measures and access systems with government agencies, companies, and open-source developers.","whyItMatters":"Anthropic is betting that restricting Mythos 5 to fixed, gated outputs lets it bring its most powerful cyber defense to a broader audience without handing attackers a prompt-driven weapon.","webCardHtml":"\u003cp\u003e\u003c/p\u003e\u003cul\u003e\n  \u003cli\u003eMythos 5 builds on Project Glasswing, an April 2026 program that gave Claude Mythos Preview to organizations protecting critical software and infrastructure.\u003c/li\u003e\n  \u003cli\u003eAnthropic says Mythos Preview found thousands of high-severity vulnerabilities in major operating systems and web browsers.\u003c/li\u003e\n  \u003cli\u003eThe public gets \u0026#34;the model 5,\u0026#34; the same base model with safety measures that block cybersecurity requests, so it cannot fully show defensive capabilities.\u003c/li\u003e\n  \u003cli\u003eAnthropic and partner companies will add misuse-prevention mechanisms so the model does not deviate from its intended use.\u003c/li\u003e\n  \u003cli\u003eThe Defender Advantage Fund is also called \u0026#34;0xDAF\u0026#34; and is worth about 5.6 billion yen.\u003c/li\u003e\n  \u003cli\u003eFor the Cyber Verification Program, defensive capabilities using Mythos-class models are planned for the future, not yet live.\u003c/li\u003e\n  \u003cli\u003eRelated reporting notes \u0026#34;the model 5\u0026#34; was temporarily halted after U.S. government export control instructions, then resumed July 1.\u003c/li\u003e\n  \u003cli\u003eFinancial Times analysis of spending data from 70,000 companies shows corporate spending on \u0026#34;the model 5\u0026#34; is flat at around 11% share, while the cheaper \u0026#34;the model 5\u0026#34; released July 24 has passed it.\u003c/li\u003e\n  \u003cli\u003eVenture firm Accel partner Miles Clements says \u0026#34;most people don\u0026#39;t need to use cutting-edge technology\u0026#34; and predicts top models may act more as showcases than profit drivers.\u003c/li\u003e\n\u003c/ul\u003e\u003cp\u003e\u003c/p\u003e","blueskyPost":"Claude Mythos 5's expanded availability still denies end users direct prompting. That design keeps the model's power inside fixed outputs, which is the actual safeguard Anthropic is selling.","twitterPost":"Claude Mythos 5 now reaches more users, but only through fixed outputs. The model never sees a direct prompt, which is the real abuse control.","threadsPost":"Claude Mythos 5's wider availability comes with a deliberate limitation: end users cannot prompt it directly. They only get fixed outputs like fix suggestions and alerts. That separation is the abuse control, and it is also what keeps the model's power out of open-ended hands.","newsletterBlurb":"Anthropic has widened access to Claude Mythos 5, its cybersecurity model, by embedding it in Claude Security with a prompt-free design. A new $35 million fund backs open-source vulnerability fixes.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20260824-claude-mythos-5-defence/\",\"title\":\"「Claude Mythos 5」の提供範囲が拡大、直接プロンプトを入力できない仕組みで悪用リスクを抑制し「Claude Security」で利用可能に＆外部サービスにも提供へ\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":13123,"outputTokens":1204,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1787890913,"createdAt":"2026-08-28T04:04:48.000Z","publishedAt":"2026-08-28T04:06:44.000Z","updatedAt":"2026-08-28T04:04:48.000Z"},"cluster":{"id":"c_64bcd358f38f12a5ed80144b","canonicalTitle":"「Claude Mythos 5」の提供範囲が拡大、直接プロンプトを入力できない仕組みで悪用リスクを抑制し「Claude Security」で利用可能に＆外部サービスにも提供へ","representativeArticleId":"a_232f39c2a2becc9f761d4171","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":false,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[],\"studios\":[],\"people\":[],\"type\":\"news\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-08-24T03:44:00.000Z","lastSeenAt":"2026-08-25T13:00:00.000Z","updatedAt":"2026-08-28T04:06:45.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20260825-anthropic-best-ai-model-struggles/","title":"Anthropicの最上位AIモデル「Claude Fable 5」は企業需要が伸びず苦戦、AI企業のビジネスモデルが覆される可能性も"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":[],"studios":[],"people":[],"type":"news","domain":"other","is_roundup":false},"keyFacts":["Claude Mythos 5 is now available in Claude Security, Anthropic's vulnerability analysis service provided as a public beta for Claude Enterprise users.","Anthropic launched the Defender Advantage Fund (0xDAF) to provide Claude usage credits worth a total of $35 million for open-source software security.","Project Glasswing launched in April 2026, providing Claude Mythos Preview to select organizations protecting critical software and infrastructure.","Claude Mythos 5 operates only for predetermined processing, with users receiving only fixed outputs like vulnerability fix suggestions and security alerts.","Anthropic's Cyber Verification Program will relax cybersecurity restrictions for approved security organizations, with Mythos-class defensive capabilities planned for the future."]}