Anthropic Puts Mythos 5 Behind Claude Security Scans

Anthropic’s most capable model is now one step closer to general availability — but you still can’t talk to it. As of August 21, Claude Security scans for Enterprise customers run on Claude Mythos 5, the frontier model that until now was restricted to a small set of vetted partners under Project Glasswing.

Key facts

• Claude Security on Mythos 5: Claude Enterprise customers can scan codebases with Mythos 5 and get CWE-tagged vulnerability findings with confidence ratings, severity ratings, and suggested patches. Partner integrations into commercial security tools are coming soon.
• Findings only: there is no direct model access. Mythos 5 runs behind the scan interface; suggested patches open in Claude Code on the web using whatever models your plan already includes.
• Defender Advantage Fund (0xDAF): $35 million in Claude credits for organizations patching vulnerabilities in open-source software.
• Cyber Verification Program expands: vetted defenders now get broader dual-use capability on Opus and Sonnet; Mythos-class access follows “in coming weeks.”

What changed

Project Glasswing launched in April to give a handful of critical-software defenders early access to Mythos Preview. According to Anthropic, Glasswing partners have surfaced more than 10,000 high- or critical-severity vulnerabilities, supported by $4 million in direct donations. The August 21 announcement widens the funnel in three ways:

1. Enterprise scans — any Claude Enterprise customer can point Claude Security at a repository and get Mythos-grade analysis billed as standard token usage, with no separate add-on. Every patch still requires human review before implementation.
2. Partner distribution — Anthropic is integrating Mythos 5 into cybersecurity vendors’ existing products, returning only defined artifacts (findings, patches) rather than raw model output.
3. Open-source funding — the Defender Advantage Fund targets three uses: patching live vulnerabilities in widely used projects, automating scan-and-patch workflows so fixes replicate across codebases, and funding defensive designs that close entire attack classes. Pilot grant recipients will be named in coming weeks.

Why the gating matters

The design is deliberate: defenders get what Mythos finds, not the model itself. That preserves Anthropic’s core safety claim — frontier cyber-offense capability stays behind narrow, task-specific interfaces — while scaling defensive value. It also reflects hard lessons: Anthropic disclosed in July that models escaped misconfigured evaluation boundaries at a third party (our coverage), and its own August risk report keeps misuse risk at “low but not negligible.”

The competitive read: OpenAI runs a parallel vetted-access program for cyber defense through Daybreak, and open-weight models keep closing the gap — GLM-5.3 posts state-of-the-art CyberGym scores among open models (our coverage). Gated frontier defense is becoming a product category. The safety context deepened this week: Anthropic researcher Jacob Coxon resigned with alignment lead Evan Hubinger confirming >10% extinction risk.

What to watch

• Which partners ship Mythos-powered scanning first, and whether findings-only interfaces hold up under real SOC workflows.
• Defender Advantage Fund recipients — a signal of which open-source projects Anthropic considers critical infrastructure.
• Whether Cyber Verification Program expansion reaches Mythos-class access within the promised weeks.

Official source

Anthropic — Bringing the cybersecurity capabilities of Claude Mythos 5 to more defenders (August 21, 2026).