Anthropic Launches Claude Mythos 5.2 With 76.3% Cybersecurity Benchmark Score to Find 0-Day Vulnerabilities

Anthropic has launched Claude Mythos 5.2, its latest AI model for cybersecurity engineering, reporting a 76.3% score on CyScenarioBench.

Released on November 9 alongside Claude Fable 5.2, the model is designed to help security teams find, reproduce, and fix software vulnerabilities. Access remains controlled because those same skills could also help attackers.

The result improves on Mythos 5.1’s 61.7%, a gain of 14.6 percentage points. However, Anthropic notes that each laboratory’s CyScenarioBench scores use different tasks.

That limits direct comparisons with competing models, making the improvement over Anthropic’s earlier release more useful than a simple ranking across vendors.

Mythos 5.2 also scored 89.6% on Anthropic’s internal CVE Rediscovery evaluation, compared with 78.3% for Mythos 5.1 and 81.3% for Opus 5.5 without safeguards.

The October 2026 test covered 1,829 published vulnerabilities, giving models project code from before each security fix and asking them to independently locate the underlying flaw.

These results need an important distinction. Rediscovering known vulnerabilities is not the same as finding new zero-day flaws unknown to software developers.

The benchmark measures performance against documented issues, rather than proving a specific success rate for unknown bugs. Anthropic says Mythos 5.2 improves vulnerability discovery, reproduction, and repair, but the launch does not establish a 76.3% zero-day detection rate.

The release follows earlier work with Mythos models on vulnerability research and code scanning. Cybersecurity News previously covered Mythos 5’s integration into Claude Security, which gave enterprise customers access to model-powered findings and suggested fixes without unrestricted access to the model itself. That separation remains central to Anthropic’s rollout.

Anthropic reports in its launch announcement that one security company found other tested frontier models missed one in six confirmed bugs identified by Mythos 5.2. Another tester reported fewer exaggerated severity ratings. These are individual testing results, not proof that the model will avoid false alarms across every codebase.

Fable 5.2 and Mythos 5.2 share the same underlying model, but their safeguards differ. Fable includes stronger restrictions for cybersecurity and biology work, while Mythos supports approved research through verification programs.

Both generate output up to 30% faster than their predecessors, with standard pricing of $8 per million input tokens and $40 per million output tokens.

The Cyber Verification Program offers Defense, Red Team, and Specialized Access tiers. Individuals can apply for Defense Access, while higher tiers require organizational approval.

Authorized penetration testing and exploit development fall under Red Team Access. Work for customers requires separate approval, and Anthropic’s usage restrictions continue to apply at every tier.

Enterprise customers can also use Mythos 5.2 through Claude Security without joining the verification program. As covered in Claude Security’s continuous scanning update, the service checks repository changes, tracks fixes, and reopens findings when flaws return.

API and MCP connections help teams move findings into existing development and security workflows.

Anthropic’s Claude Security documentation keeps human review central. Teams must review and approve proposed patches, especially for critical systems.

Faster discovery can shorten investigations, but finding a flaw is only the beginning. Security teams still need to validate its impact and test the fix.

The post Anthropic Launches Claude Mythos 5.2 With 76.3% Cybersecurity Benchmark Score to Find 0-Day Vulnerabilities appeared first on Cyber Security News.