https://www.anthropic.com/research/glasswing-initial-update
Anthropic appears to be preparing a broader rollout of Claude Mythos. First announced in early April as a restricted early preview, Mythos represents a significant capability jump over Anthropic’s current flagship model, Opus 4.7, particularly in the area of autonomous cyber operations. Anthropic has stated that the model can automatically develop functional cyberattacks at a highly professional level and explicitly warned that in the short term, an unrestricted release could advantage attackers over defenders if frontier AI laboratories are not careful about how they bring such models to market. The company chose not to release Mythos publicly until it had developed a guardrail system capable of managing those risks, and recent evidence suggests that system may now be in place.
References to a model identifier labelled claude-mythos-1-preview have appeared in both Claude Code and Claude Security, Anthropic’s developer-facing tools, and at least one user briefly observed a toggle to enable the Mythos model within the public version of Claude Code before it was taken offline. It remains unclear whether the model will be available across all subscription tiers or restricted to specific use cases and user categories.
The appearance of Mythos references in Claude Code and Claude Security suggests that Anthropic is moving toward making a model with professional-grade offensive cyber capabilities available to a broad population of developers and security practitioners. A shift that, if managed well, represents a substantial uplift for defensive security operations, but one that also compresses the timeline within which organisations need to have mature AI security governance in place.