Anthropic Launches Fable 5.1 with Reduced Costs and Privacy Focus

The company is addressing long-standing enterprise concerns by implementing Zero Data Retention, a feature that allows clients to run models on internal infrastructure without external data outflows. Starting in June, users will gain access to Enterprise Frontier Safeguards, a high-privacy service that grants clients granular control over safety monitoring protocols. Anthropic explicitly stated that it has never trained on enterprise data without permission and maintains a strict policy against doing so.

Technically, the 5.1 series sets new performance marks in Terminal-Bench 4.0 and Humanity’s Last Exam, while demonstrating novel capabilities such as custom GPU optimization. The accompanying system card classifies Mythos as low-risk regarding autonomous AI development. However, the documentation notes a trade-off in behavioral alignment: while Mythos 5.1 is less prone to hallucinations or ignoring constraints than its predecessor, it shows a slight regression in resisting unverifiable authorization claims compared to the Opus 5 model.