Bringing the cybersecurity capabilities of Claude Mythos 5 to more defenders

Anthropic first gave select organizations access to Claude Mythos Preview (and its successor, Claude Mythos 5) through Project Glasswing in April, letting defenders find and fix vulnerabilities before similarly capable models went general or reached malicious actors. To expand this safely, the company developed safety classifiers and safeguards; Claude Fable 5 was the first step, making a model broadly available while blocking dual-use cyber work.

The next step reframes the risk model: direct model access is risky because users can steer it toward harmful uses, but delivering only specific outputs like vulnerability patches or security alerts carries far less risk. Accordingly, Anthropic is integrating Claude Mythos 5 into partners' cyber defense tools, making Claude Mythos 5 available for codebase scans in Claude Security on Enterprise plans, launching the Defender Advantage Fund (0xDAF) with $35 million in credits for organizations patching open-source projects and automating scan/patch processes, and expanding the Cyber Verification Program to include broader dual-use capabilities on Opus and Sonnet (with Mythos-class access to follow).

The aim is to help organizations adapt to the pace of cybersecurity as AI models become more powerful, while continuing to develop safeguards, access programs, and community support to safely make frontier models widely available to defenders.