Anthropic launches Claude Mythos 5.2 for finding and fixing software vulnerabilities
- Severity
- Info
- Sources
- 1
What happened
Anthropic's Claude Mythos 5.2 scored 76.3% on CyScenarioBench (up from 61.7% for Mythos 5.1) and 89.6% on its internal CVE Rediscovery evaluation of 1,829 published vulnerabilities, where models received pre-fix code and had to locate the flaw. CyScenarioBench tasks differ between vendors, so cross-vendor comparison is limited, and the results do not establish a zero-day detection rate. Access is gated through the Cyber Verification Program, and Claude Security offers enterprise access without joining it. Defenders should keep human review of proposed patches, especially for critical systems, and validate impact and test fixes before acting on AI findings.
For security analysts · Written by AI from 1 source · Wrong?
What to do
- Keep human review of any AI-proposed patches, especially for critical systems.
- Validate the impact of AI-reported flaws and test the proposed fix before deploying it.
- Individuals wanting access must apply through the Cyber Verification Program, while higher tiers need organisational approval.
Written by AI from the sources, checked against them
Full story timeline · 1 update
Who reported · 1 outlet
Who and what is in it
Companies