What happened
On Aug 21, 2026 Anthropic enabled Claude Security (public beta for Claude Enterprise) to run scans on Claude Mythos 5 — its most cyber-capable, previously access-restricted model — returning CWE-classified findings and suggested patches without exposing the model itself; Anthropic also launched a $35M Defender Advantage Fund (0xDAF) for open-source security and previewed expanding its Cyber Verification Program and partner integrations (Wiz, Palo Alto, CrowdStrike, Accenture already build on Claude for cyber).
Why it matters
This is the first time Anthropic's most powerful cyber-reasoning model is available to any Enterprise customer (not just vetted Project Glasswing partners), via an output-only access pattern that could become the template for safely commercializing dual-use frontier capability in security tooling.
Applicability
Enterprise security teams with Claude Enterprise agreements and GitHub-hosted code should evaluate Claude Security now for pre-release audits and legacy code scanning; security vendors should evaluate the Mythos partner-integration program.