When AI Finds Bugs Faster Than Teams Can Patch Them
Introduction The most compelling AI story emerging in 2026 isn’t about chatbots, image generation, or code autocompletion. It comes from the defensive security space. In April 2026, Anthropic launched Project Glasswing, granting technology leaders and critical infrastructure providers early access to Claude Mythos Preview — a specialized security model built for vulnerability discovery and defensive analysis. The findings were stark: out of an estimated 6,202 high- or critical-severity vulnerabilities identified across foundational open-source software, only 97 had been confirmed remediated within two months. Anthropic noted that “even at our relatively slow pace of disclosures, Mythos Preview is adding to an already-overloaded security ecosystem.” ...