Claude Mythos Preview: Anthropic's Restricted Advanced Model

Anthropic released Claude Mythos Preview, a model the company describes as a 'fundamentally new model class' optimized for cybersecurity, coding, and complex reasoning. Anthropic deliberately did not make it generally available. Access is restricted to select internet-critical companies and open-source maintainers via Amazon Bedrock (gated research preview) and Vertex AI. Anthropic used Mythos Preview to find thousands of zero-day vulnerabilities across every major OS and browser. In response, they launched Project Glasswing — a coalition of 40+ organizations (AWS, Apple, Google, Microsoft, NVIDIA, Cisco, CrowdStrike, JPMorganChase, Linux Foundation, Palo Alto Networks) to help secure critical software. Separately, the UK AI Safety Institute published an evaluation of Mythos' cybersecurity capabilities.

For AI GTM operators and buyers: Mythos signals that frontier capability is ahead of what's publicly available. This is the second major model Anthropic has deliberately restricted — after early Opus 4.6 capacity constraints. For buyers evaluating AI vendor roadmaps: "what capabilities are in your restricted models and when will they reach GTM/sales/marketing use cases?" is now a legitimate procurement question. For content writers: the "responsible release" vs. "racing" dynamic is the real story behind the Anthropic brand positioning.

• "Anthropic built a model so capable they won't release it. Here's what that means for your AI stack"
• "Claude Mythos found zero-days in every major OS. The model you're using is the 'safe' version."
• "The Mythos release pattern: Anthropic is deliberately sequencing capability vs. access"
• "What Project Glasswing means: AI's first industry-wide defensive coordination effort"