The 10,000-Bug AI: How Anthropic’s Secret “Mythos” Model is Rewriting Cyber-Resilience

Anthropic's Project Glasswing uses a high-capability internal model, Claude Mythos Preview, to dramatically accelerate vulnerability discovery—claiming discovery of over 10,000 unique defects (many critical/high-risk) across enterprise and open-source targets—and is being tightly restricted and partnered with major vendors and governments due to the risk that exfiltration of the model could enable automated, large-scale zero-day exploitation.