How AI is getting better at finding security holes

Anthropic announced Mythos Preview, an advanced foundation model that the company says can find high-severity vulnerabilities across major operating systems and browsers; access is being tightly limited under "Project Glasswing." The article describes how AI-driven bug-finding has recently improved, helping maintainers find more legitimate security issues while raising concerns that more capable or open-weight models could be weaponized to create exploit code, with defenders still uncertain about AI's ability to reliably fix complex bugs.