Anthropic's AI Finds 23,000 Flaws in Critical Software Projects

Anthropic is expanding Project Glasswing, a collaborative initiative aimed at enhancing cybersecurity for critical software using AI. Initially begun with 50 partners, it now includes 150 more organizations across 15 countries, including significant players in power, healthcare, and communications sectors. The program utilizes Claude Mythos, a tool that has identified over 23,000 potential vulnerabilities across numerous codebases.

Although many flaws were discovered, only a small fraction have been patched, highlighting a challenge in verifying and addressing these vulnerabilities. Anthropic is also focused on improving the patching process and sharing best practices for dealing with open-source vulnerabilities.