Project Glasswing finds 10,000 critical flaws in month

AI-driven vulnerability discovery has outpaced human patching capacity, shifting the security bottleneck entirely to remediation.

• –The Claude Mythos model found a 27-year-old flaw in OpenBSD and a 16-year-old flaw in FFmpeg that survived decades of human audits
• –Anthropic is withholding general release of the model to prevent asymmetric attacker advantages, sharing it only with 50 vetted partners
• –The sheer volume of discoveries highlights an urgent need for automated patching workflows, as human maintainers cannot triage this many issues
• –Alongside the update, Anthropic launched Claude Security in beta to help enterprise teams scan for issues during the commit process