Anthropic report says "Claude Mythos" found over 10,000 vulnerabilities, but fixes couldn't keep up

Anthropic said its Claude-based Project Glasswing and open-source scans uncovered more than 10,000 high- and critical-severity vulnerabilities.

The company also reported over 1,600 disclosures from its own scans, with most findings still unpatched.

The gap between rapid AI-driven discovery and slower remediation is emerging as a major security bottleneck.

To support enterprise use, Anthropic launched Claude Security and a new Cyber Verification Program for researchers.