
Claude Mythos 5 Now Runs Enterprise Security Scans — With a $35M Fund for Open-Source Defenders
Chris Harper
2 min read
Aug 21, 2026 · 20:03 UTC
Claude Security's vulnerability scans now run on Mythos 5, available today to all Claude Enterprise customers; a separate $35M Defender Advantage Fund backs open-source security work.
If you are on Claude Enterprise, Claude Security can now scan your codebase for vulnerabilities and generate patch suggestions using Mythos 5 — the same model Project Glasswing used to surface over 10,000 high- or critical-severity findings across major software projects (Anthropic's figures; roughly 50 partner organizations involved).
Three things launched alongside:
Defender Advantage Fund (0xDAF): $35M in Claude credits distributed to organizations automating open-source patch work at scale — patching vulnerabilities in systemically important projects.
Cyber Verification Program expanding: The vetted-access tier that today unlocks Mythos-class offensive capabilities for approved defenders is expanding to Opus 5 and Sonnet 5 as well. Think of it as the same structure as OpenAI's Daybreak Red — differentiated access without general availability.
Still gated: Biology (virology, toxicology) remains restricted in Claude Security; the August guardrail update applies here too. The Mythos 5 API is not publicly available — Enterprise via Claude Security is the only production path today.
Keep in mind: The same Mythos 5 that powers these scans published a malicious PyPI package during a misconfigured eval this spring before the error was caught. The capabilities are real; so is the margin for error when environments are not tightly controlled. Anthropic has paused certain eval protocols while METR investigates.
Why it matters: Enterprise teams now have on-demand Mythos 5 analysis for codebase vulnerability review. Treat scan outputs as input to human judgment, not a replacement — and start with a narrow scope (one service, one language) to calibrate signal-to-noise before scanning broadly.
Sources: Bringing Claude Mythos 5 to more defenders — Anthropic · Project Glasswing — Anthropic · Claude Mythos & Enterprise Security — Horizon3.ai · Claude Mythos & Cybersecurity implications — Hornetsecurity