Anthropic AI scanner finds 29,000 open source bugs
Anthropic's automated tool, powered by its Claude model, identified over 29,000 potential vulnerabilities across major open-source software projects in six months. Human security researchers have only reviewed about 6,000 of these, confirming most as valid, but less than 600 have been patched by maintainers. Anthropic now offers direct scans and unvalidated reports to maintainers, bypassing slower human review.
Why it mattersThe scale at which Anthropic’s AI uncovers vulnerabilities shows the advantage of automated tools in security research but also demonstrates how human triage and industry response lag significantly behind. This gap leaves many potential security issues unresolved in publicly used open-source software.
- Claude scanner found 29,000 possible open source vulnerabilities
- Human teams reviewed only 6,123, confirmed 5,674 as valid
- Just 516 vulnerabilities have been patched to date
- Anthropic now sends unreviewed reports to maintainers upon request
- Early tests show most critical findings were legitimate
Sources covering this
In this story
More in Cybersecurity
Malware exposes over 8 million karaoke customer records
A malware infection at Nippon Columbia, a contractor for Japanese karaoke giant Daiichi Kosho, potentially exposed personal data of more…
GTA 6 leaker threatens early release over demands
A leaker calling themselves Cyberleek claims to have a working copy of GTA 6 and is threatening to leak the full story mode online…