Anthropic AI Finds 10,000+ Software Vulnerabilities

Anthropic's Project Glasswing has uncovered over 10,000 software vulnerabilities using AI. This includes thousands of zero-day flaws that had gone undetected for years, even decades. Here's the thing: An unreleased AI model, Claude Mythos Preview, found these issues in just weeks. For example, it identified a 27-year-old vulnerability in OpenBSD and a 16-year-old bug in FFmpeg. These bugs had previously survived extensive human review and automated testing. What's interesting is that while over 10,000 high- and critical-severity vulnerabilities have been identified, fewer than 100 have been patched so far. This shows the AI can find flaws much faster than they can be fixed. The bottom line: This initiative highlights the significant potential of AI in cybersecurity, but also reveals a major bottleneck in the process of fixing these newly discovered flaws.