Anthropic’s AI Exposes Bugs Faster Than Microsoft Can Patch Them

In mid-May, Microsoft engineers met at the company’s Redmond headquarters to discuss Project Glasswing, an effort to address code vulnerabilities uncovered by Anthropic’s new AI model, Mythos. Anthropic had given select organizations access to the tool so they could find and fix weaknesses before hackers or adversarial governments could exploit them. During the meeting, engineers questioned whether Mythos lived up to Anthropic’s claims.