Anthropic's Claude finds 10,000 code flaws in 30 days

Anthropic's Project Glasswing has surfaced more than 10,000 critical vulnerabilities across production codebases in a single month, a scale of automated security auditing that dwarfs anything previously reported in the industry. The initiative uses Claude as its core reasoning engine, running it against real-world repositories rather than sandboxed test environments. Anthropic claims a zero false-positive rate across all findings, which, if it holds up to independent scrutiny, would represent a meaningful leap over conventional static analysis tools that typically generate substantial noise alongside real detections. Essentially: Anthropic is using Claude to do the work that thousands of security researchers would take years to complete, and doing it with claimed precision that commercial tooling hasn't matched. - 10,000+ critical flaws verified in one month, all in production codebases, none dismissed as false positives - Scale significantly exceeds prior Glasswing disclosures, suggesting the project has expanded its scope or Claude's capability has materially improved - Zero false positives is the headline claim that will face the most scrutiny from the security research community If the numbers hold, this is less a proof-of-concept than an operational argument that AI-assisted security auditing is ready to replace, not just supplement, traditional penetration testing pipelines.