Anthropic is advancing its cybersecurity offerings by enhancing access to its sophisticated AI models through new collaborations and updates. The initiative includes an expanded Claude Security service, an innovative open source funding scheme, and plans to grow its Cyber Verification Program.
Expanding Mythos 5 Access
Building on the foundation of Project Glasswing, launched in April, Anthropic is increasing availability of its Claude Mythos 5 model. Initially, a select group of organizations were granted early access to Claude Mythos Preview and Mythos 5 to identify and address vulnerabilities before these capabilities became widely available. The broader rollout of Claude Fable 5 ensures dual-use cyber functionalities remain restricted.
Anthropic emphasizes that the greatest risk arises from unrestricted access to AI models. By providing specific defensive outputs such as patches or alerts, they aim to significantly reduce this risk while enhancing what defenders can achieve with Mythos-class models.
Strategic Integrations with Cybersecurity Partners
Collaborating with cybersecurity firms, Anthropic is embedding Mythos 5 into existing security tools. This integration is designed to support teams safeguarding key sectors such as healthcare, utilities, financial systems, and software supply chains.
End users interact with the AI through specialized interfaces, receiving only necessary outputs like suggested patches. These interfaces incorporate abuse-prevention mechanisms, ensuring the model remains within its intended use scope.
Currently in public beta, Claude Security utilizes Mythos 5 for codebase scanning. The scans provide findings categorized by CWE, along with ratings for confidence and severity, and recommend fixes. Each fix requires implementation via Claude Code and human approval before deployment.
Introducing the Defender Advantage Fund
Anthropic has launched the Defender Advantage Fund, allocating $35 million in Claude credits to assist open source maintainers in securing their projects. This initiative follows $4 million in previous donations and support through Project Glasswing.
The fund will be used to address live vulnerabilities, develop reusable scanning and patching processes, and explore security methodologies to counter entire attack classes. Initial investments will be focused on a few substantial pilot grants.
The Cyber Verification Program, already offering reduced safeguards for authorized work on Claude Opus and Sonnet, is set to expand. In the coming weeks, it will include broader capabilities on these models, such as vulnerability triaging, with Mythos-class access enhancements to follow.
Anthropic continues to extend Mythos access through Project Glasswing, in collaboration with U.S. government partners. This effort is particularly aimed at organizations that protect critical infrastructure while meeting stringent security control standards.
Security teams are encouraged to apply to the Cyber Verification Program to gain reduced safeguards on Opus and Sonnet, with more details on the extended rollout anticipated soon.
