Anthropic shelves Mythos Preview, will test safeguards on upcoming Claude Opus
Anthropic said it will not release Mythos Preview and plans to trial its safety controls on a forthcoming Claude Opus model.
Anthropic announced on April 7 that it will not make Mythos Preview generally available, citing the need for safeguards that reliably block the model class's most dangerous outputs before any wider deployment. The company said it will begin testing those safeguards with an upcoming Claude Opus model, according to a post on X from the official Anthropic account.
The decision signals that Anthropic is treating Mythos-class models as a capability tier requiring new controls rather than incremental guardrails, and that Claude Opus will serve as the proving ground for those mechanisms. For teams building on Anthropic's API, the practical implication is that access to the most capable frontier tier remains gated behind safety work, while near-term product roadmaps will continue to center on the Claude Opus line. Buyers evaluating vendor risk now have a clearer statement of Anthropic's deployment posture: capability releases are conditional on safeguard readiness, not schedule.