Anthropic expanded Project Glasswing on 3 June 2026 — giving 150 new organisations across 15 countries access to Claude Mythos Preview, its most powerful cybersecurity AI. The first 50 partners found 10,000 high-severity flaws. The new cohort covers power, water, healthcare, and communications infrastructure. Anthropic warned a successful attack on most partners could affect 100 million people or more.