Anthropic widens Mythos 5 access, backs open-source with $35M

Anthropic announced expanded access to its Mythos 5 model for security teams, integrating the model into partner tools, updating Claude Security to run on Mythos 5, enlarging its Cyber Verification Program and creating a $35 million Defender Advantage Fund in Claude credits for open-source security work.

The company said the integrations allow security operations, incident response and detection platforms to run Mythos 5 behind purpose-built interfaces. End users will receive defined outputs such as suggested patches or alerts rather than direct access to the model. These interfaces include abuse-prevention checks to limit harmful or dual-use outputs.

Claude Security, now in public beta for Claude Enterprise customers, runs its codebase scans on Mythos 5. Scans list each finding with a Common Weakness Enumeration category, confidence and severity ratings, and a suggested fix. Any fix must be implemented through Claude Code and approved by a human before deployment. Anthropic explained, “Claude Security uses Mythos 5 to scan code you own, and returns detailed findings rather than raw outputs without exposing the model itself.”

The Defender Advantage Fund will distribute $35 million in Claude credits to organizations that help open-source maintainers find and patch vulnerabilities, build reusable scanning and patching processes, and develop defenses that resist whole classes of attack. Anthropic said the initial grants will focus on a small number of larger pilot awards for urgent patching and reusable tooling. The fund follows about $4 million in prior donations and coordinated efforts from Project Glasswing.

The Cyber Verification Program, which currently provides vetted organizations reduced safeguards on Claude Opus and Sonnet for authorized security work, will expand to cover broader dual-use capabilities on those models, including vulnerability triage and validation. Anthropic plans to extend Mythos-class access through the program in the coming weeks. The company is continuing Project Glasswing with U.S. government partners to give Mythos access to organizations that protect critical infrastructure and meet strict security controls.

Anthropic said it views direct, unrestricted access to a powerful model as the highest risk and that providing specific defensive outputs reduces that risk. Integrations and product limits are designed to return only defined outputs and to include checks that prevent production of harmful instructions.

Security teams seeking reduced safeguards on Opus and Sonnet can apply to the Cyber Verification Program now. Anthropic said it will publish additional details about the broader rollout and grant process in the coming weeks.