Anthropic AI Finds 10,000 High-Severity Flaws in Critical Software

Anthropic’s Project Glasswing, an AI-driven cybersecurity initiative, has reportedly uncovered over 10,000 high- or critical-severity vulnerabilities in globally significant software. The Hacker News reports that this effort, which launched last month, involves a small consortium of around 50 partners leveraging Anthropic’s AI to identify these flaws.

This isn’t just about finding bugs; it’s about the scale and the implications. Ten thousand severe vulnerabilities in “systemically important” software is a massive haul in a short period. It signals that even mature, widely used codebases are riddled with exploitable weaknesses that traditional security testing might be missing or struggling to keep pace with.

For defenders, this highlights a critical reality: our attack surface is far more exposed than many realize. Attackers are constantly probing these same codebases, and if an AI can find this many flaws so quickly, you can bet threat actors are already exploiting a subset of them. CISOs must consider the implications for their supply chain and the third-party software they rely on. The attacker’s calculus here is simple: more vulnerabilities mean more opportunities for initial access.