On Tuesday, Anthropic introduced Fable and Mythos 5.1, two paired versions of its most sophisticated AI model to date. Beyond enhanced performance, the latest Fable iteration brings adjustments aimed at lowering token costs and reducing false-positive blocks from the model’s safety filters. As with the earlier Mythos release, Mythos 5.1 remains exclusive to approved Anthropic collaborators working in cybersecurity or life sciences. The open version, Fable 5.1, becomes available today through cloud services or the Anthropic API.

One of the standout updates is Anthropic’s previously mentioned adoption of Zero Data Retention, which lets customers run Anthropic models on their own infrastructure without any data flowing back to the company. A high-privacy option, dubbed Enterprise Frontier Safeguards, was previously unavailable for Fable due to security concerns but will now be offered to users starting in June. Notably, the system will still watch for misuse by both automated agents and human users, though clients will have control over how that monitoring is carried out. In the announcement, Anthropic reassured customers that their information had not been improperly accessed, stating, “Anthropic has never trained on enterprise data without explicit permission, and never will.”

As typical for an Anthropic launch, the new models have set records across numerous benchmarks, including Terminal-Bench 4.0 for command-line coding and Humanity’s Last Exam for general reasoning. Anthropic also shared three fresh scientific discoveries produced by the models prior to release, such as a custom GPU optimization and a high-resolution map of Venus built from existing imagery. Like earlier versions, these models arrive with a comprehensive system card that explains their capabilities in straightforward terms. The card rates Mythos as “low-risk” regarding automated AI development—where the AI improves itself—a factor some view as a potential precursor to losing human oversight. It notes, “its ability to accelerate internal AI R&D progress is in line with current trends.”

When it comes to general misbehavior, Mythos shows a slight uptick compared to Opus, likely stemming from its expanded abilities. The system card states, “Mythos 5.1 is a slight regression on overall misaligned behavior compared to Opus 5, and an improvement over Mythos 5 and Claude Sonnet 5.” It adds, “It cooperates with human misuse and accepts unverifiable claims of authorization somewhat more readily than Opus 5, but it is less likely to ignore explicit constraints, hallucinate inputs, or falsely claim to have completed tasks than previous models.”