Anthropic appears to be preparing a public rollout of its restricted Claude Mythos model, which it said in April could create highly capable cyberattacks and expose private and public software to major security risks. The model surfaced briefly in Claude Code and Claude Security, including as a preview model announcement under the name claude-mythos-1-preview.
KEY FACTS
- Model name The preview version is called claude-mythos-1-preview.
- Public signs Users briefly saw a toggle to enable Mythos in Claude Code before it was removed.
- Security focus Anthropic said the model made major gains in code reasoning and autonomy.
- Current use The model is being used in Glasswing, a project focused on finding AI-driven exploits.
The disclosure said Mythos can automatically develop functional cyberattacks at a highly professional level. Anthropic delayed a wider release until it could build guardrails strong enough to reduce the risk that attackers could use it against unpatched software.
The company has also said it is working with outside firms on Glasswing, a project that uses the unreleased Claude Mythos Preview to help secure critical software. Anthropic said the effort has already helped up to 50 organizational partners and that Mythos found 10,000 high- or critical-severity vulnerabilities in its first month.
It remains unclear whether the model will be offered across all subscription tiers if it reaches the public version of Claude Code or Claude Security. Anthropic currently sells Claude Opus 4.7, Opus 4.6, Opus 4.5, Sonnet 4.6, and Haiku 5.5.
WHY IT MATTERS
A broader release of a model that can help generate advanced exploits could affect both attackers and defenders. The company says stronger controls are needed before public access, but the model’s security findings also suggest it may be useful for finding flaws before software ships.