What Happened
On 2026-04-07, Anthropic announced Claude Mythos Preview and Project Glasswing, a coalition giving selected infrastructure organizations access for defensive cybersecurity.
Why It Matters
This was a consequential deployment decision: a highly capable general-purpose model was kept out of general availability because of its cybersecurity capabilities. The event therefore belongs in both the capability and deployment history of AI. It illustrates how access policy can become part of a frontier-model launch, rather than a separate product decision made afterward.
Technical Details
Launch partners included major cloud, software, and infrastructure organizations. Permitted work covered finding and fixing vulnerabilities in critical systems. Anthropic published technical security evaluations; its claims about discovered vulnerabilities should remain attributed to the company.