Anthropic brings Mythos 5 into controlled cyber workflows<\/p>\n
Summary<\/p>\n
Anthropic is bringing Claude Mythos 5 into security products and committing US$35 million in credits for open-source vulnerability remediation.<\/p>\n
Anthropic is expanding the defensive use of Claude Mythos 5, its restricted frontier model for cybersecurity, by bringing it into security products and services while maintaining safeguards around direct model access.<\/p>\n
Announced on 21 August, the move includes Claude Security scans powered by Mythos 5 for Claude Enterprise customers, planned integrations with cybersecurity technology and services providers, and a US$35 million credit fund for open-source security work.<\/p>\n
Anthropic also plans to broaden its Cyber Verification Program, which gives vetted organisations access to certain dual-use cybersecurity capabilities under controlled conditions.<\/p>\n
Claude Mythos 5 comes to Claude Security<\/p>\n
Anthropic says Claude Security can now use Mythos 5 to scan customer codebases for security vulnerabilities and suggest patches.<\/p>\n
The capability is available in public beta to Claude Enterprise customers. Enterprise administrators can enable Claude Security from the admin console, after which users can select a repository for scanning.<\/p>\n
The scans return detailed findings, including:<\/p>\n
Users can then use Claude Code on the web to implement suggested fixes. Anthropic says every patch must be reviewed and approved by a human before it is implemented.<\/p>\n
The company has separated defensive use of Mythos 5 from direct, unrestricted access to the underlying model. The model performs the security analysis, while users receive the resulting findings and recommended fixes through Claude Security.<\/p>\n
Anthropic is embedding Mythos 5 in security tools<\/p>\n
Anthropic says security teams already rely on established products and services for security operations, incident response, threat intelligence and detection engineering.<\/p>\n
Rather than requiring teams to use a standalone model interface, the company plans to integrate Mythos 5 into the tools and services they already use.<\/p>\n
The model could, for example, analyse a vulnerability within a remediation product and return suggested patches. The end user receives the remediation output without being able to directly prompt the underlying model to develop an exploit.<\/p>\n
Anthropic says it and its partners are implementing measures to prevent abuse and keep such integrations within their intended defensive scope. The company is inviting cybersecurity product and service providers to register interest in integrating Mythos 5.<\/p>\n
US$35 million fund targets open-source security<\/p>\n
Anthropic is committing US$35 million in Claude credits through its new Defender Advantage Fund, or 0xDAF.<\/p>\n
The programme is intended to support organisations that help open-source maintainers improve the security of widely used software. Grants will focus on three areas:<\/p>\n
Patching live vulnerabilities in widely used open-source projects<\/p>\n
Automating vulnerability scanning and patching so the approach can be replicated across projects<\/p>\n
Developing security approaches that make projects more resilient to entire classes of attacks<\/p>\n
The fund builds on Project Glasswing, through which Anthropic committed US$4 million in direct donations to open-source security organisations, alongside usage credits and support for vulnerability scanning and remediation.<\/p>\n
Anthropic says it will begin with a small number of larger pilot grants before expanding the programme.<\/p>\n
Cyber Verification Program gets broader access<\/p>\n
Anthropic is also expanding its Cyber Verification Program, which allows vetted organisations to use certain dual-use cybersecurity capabilities under reduced safeguards for legitimate defensive work.<\/p>\n
The programme currently covers select Claude Opus and Sonnet models. Over the coming weeks, Anthropic plans to expand safeguarded access to broader dual-use capabilities, with controlled Mythos-class access expected to follow.<\/p>\n
Approved cyber defenders will gain greater scope for tasks such as vulnerability triage and validation while continuing to operate within the programme\u2019s safeguards.<\/p>\n
Anthropic also says it will continue expanding Mythos access through Project Glasswing, including in collaboration with US government partners and organisations protecting critical infrastructure that meet strict security requirements.<\/p>\n
Separating defensive outcomes from model access<\/p>\n
The larger strategy is to make increasingly capable AI useful to cybersecurity teams without providing unrestricted access to capabilities that could also be used offensively.<\/p>\n
Anthropic argues that risk is greatest when a user can directly interact with a powerful model and attempt to steer it towards harmful activity. A security product, by contrast, can expose a narrower output, such as a vulnerability report or suggested patch, while keeping the underlying model behind additional controls.<\/p>\n
\u201cThe riskiest behavior occurs when a user has direct access to a model, where a malicious actor can try to steer it toward harmful uses,\u201d Anthropic said.<\/p>\n
The approach combines frontier-model capabilities with product-level restrictions, human review and controlled access programmes.<\/p>\n
What this means for enterprise cybersecurity<\/p>\n
The announcement points to a wider shift in how frontier AI models may be deployed for cybersecurity. Rather than providing every security professional with direct access to a highly capable general-purpose model, Anthropic is embedding its capabilities in purpose-built and constrained workflows.<\/p>\n
For enterprise security teams, that could mean using Mythos 5 for vulnerability discovery, triage, validation and remediation without changing their established workflows or exposing the full model to end users.<\/p>\n
For open-source projects, the Defender Advantage Fund creates another route for applying AI to the persistent challenge of identifying and fixing vulnerabilities across widely used software.<\/p>\n
Anthropic says the initiatives are intended to help defenders respond to the increasing pace of AI development and strengthen the security of critical software infrastructure.<\/p>\n","protected":false},"excerpt":{"rendered":"Anthropic brings Mythos 5 into controlled cyber workflows Summary Anthropic is bringing Claude Mythos 5 into security products…\n","protected":false},"author":2,"featured_media":149042,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8],"tags":[288,4989,53,3154,182,38060,16187,21173,22137,313,73007,523,22908,47056],"class_list":["post-149041","post","type-post","status-publish","format-standard","has-post-thumbnail","category-anthropic","tag-ai-cybersecurity","tag-ai-safety","tag-anthropic","tag-anthropic-claude","tag-claude","tag-claude-mythos-5","tag-claude-security","tag-cyber-defence","tag-cyber-verification-program","tag-cybersecurity","tag-defender-advantage-fund","tag-enterprise-ai","tag-open-source-security","tag-vulnerability-detection"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/149041","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=149041"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/149041\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/149042"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=149041"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=149041"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=149041"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}