Anthropic model is finding Microsoft code bugs faster than engineers can fix them
Ars Technica AI · rss · 2026-07-29
Anthropic's AI model is reportedly finding vulnerabilities faster than Microsoft can patch them.
According to Ars Technica, Microsoft engineers met in May to review Project Glasswing after a new Anthropic model uncovered weaknesses in Microsoft code at an unusually fast rate. The article says Anthropic had given select organizations access to the model so they could find and fix flaws before attackers or hostile states could exploit similar tools for espionage or sabotage.
More from Safety
- AI Safety Researchers Debate Why Model Goal Guarding Mechanisms Fail — RyanGreenblatt · 2026-07-30
- US Commits $5B to Genesis Mission, A National AI for Science Moonshot — PeterDiamandis · 2026-07-30
- OpenAI Autonomous Models Compromised Hugging Face Credentials During Security Eval — The Decoder · 2026-07-30
- Google ADK article argues tool-call interceptors are the missing safety layer for agents — fhinkel · 2026-07-30
- AI Safety Researcher to OpenAI: 'Entropy Doesn't Negotiate' — basedjensen · 2026-07-30
- AI governance could spiral out of control as models outpace institutions — LuizaJarovsky · 2026-07-30