What happened
Project Glasswing is not just another benchmark story. Anthropic says roughly 50 partners have used Claude Mythos Preview, and that most partners have each found hundreds of critical or high-severity vulnerabilities. Collectively, the partners have found more than ten thousand.
The bottleneck has moved. It used to be finding the bugs. Now it is verifying, disclosing, patching, and deploying fixes fast enough that defenders keep the advantage.
That is the executive takeaway: AI security is becoming an operations problem, not only a research problem.
The board-level question is simple: if AI can find vulnerabilities this quickly, can your organization patch at the same speed?
Source
Reported by Project Glasswing: An initial update via anthropic.com, published May 22, 2026.