Anthropic Withholds Claude Mythos from Public, Launches Project Glasswing for Cyber Defense

Anthropic has taken the unusual step of withholding its most capable model, Claude Mythos Preview, from a general public release due to cybersecurity concerns. Instead, the company announced Project Glasswing on April 7, 2026 — a controlled initiative that gives select organizations early access to Mythos specifically for defensive security work on critical software infrastructure.

The list of participating companies reads like a who's who of the technology industry: Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, and NVIDIA are among the founding members, alongside roughly 40 additional organizations responsible for building or maintaining the world's most widely used software. The goal is to use Mythos's unprecedented coding and reasoning capabilities to find and fix vulnerabilities before malicious actors can exploit a similarly capable model to discover them first.

The stakes are significant. Over recent weeks, Anthropic used Claude Mythos Preview internally to identify thousands of zero-day vulnerabilities across every major operating system and every major web browser. The scope of those findings — which have not been publicly disclosed — is reportedly what convinced Anthropic that a standard public launch would be premature and potentially dangerous.

Access to Mythos Preview under Project Glasswing is priced at $25 per million input tokens and $125 per million output tokens, available via the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. Anthropic has committed $100 million in model usage credits to the program to support the research preview. The company says Project Glasswing represents a new model for how frontier AI capabilities can be responsibly introduced to the world — starting with defense before offense.