Anthropic’s AI finds 10,000 high flaws, patching lags behind

Anthropic's Project Glasswing has discovered over 10,000 high- or critical-severity vulnerabilities since its launch, highlighting a significant gap in patching capabilities in cybersecurity. This initiative, involving collaboration with major tech firms, aims to use AI (Claude Mythos Preview) to systematically identify and address software flaws before they can be exploited. Of the vulnerabilities identified, 1,726 were confirmed exploitable, with 1,094 rated as serious.

The project emphasizes the growing disparity between the speed at which vulnerabilities are found and the slower pace of patches, posing a structural challenge in cybersecurity. Anthropic is cautious about releasing powerful AI tools to the public due to potential misuse risks.