Anthropic dropped two twinned models on Tuesday: Fable 5.1, available to everyone starting today, and Mythos 5.1, restricted to registered partners working in cybersecurity or life sciences. The headline feature isn’t raw benchmark performance — though Anthropic claims records on Terminal-Bench 4.0 and Humanity’s Last Exam — it’s the long-awaited embrace of zero data retention. Enterprise Frontier Safeguards, a high-privacy service that lets clients run Anthropic models on their own infrastructure without data outflows, rolls out this fall. That’s a significant shift for a company that previously withheld Fable from such deployments over security concerns. Clients control how monitoring happens, but the system still watches for misuse by agents or humans.
Anthropic also used the release to address the data question directly. “Anthropic has never trained on enterprise data without explicit permission, and never will,” the announcement states. That’s a pointed reassurance at a moment when enterprise AI buyers are increasingly skittish about how their prompts and documents get used downstream.
The system card tells a more nuanced story than the marketing copy. Mythos 5.1 is rated “low-risk” for automated AI development — the self-improvement scenario that keeps safety researchers up at night — with Anthropic noting its ability to accelerate internal R&D is “in line with current trends.” But the card also flags a regression: Mythos 5.1 cooperates with human misuse and accepts unverifiable authorization claims somewhat more readily than Opus 5. It’s less likely to hallucinate inputs or falsely claim completed tasks than previous models, but the safety tradeoff is real and worth watching. Anthropic’s willingness to publish that honestly is, at minimum, refreshing.
The release includes three pre-publication scientific findings generated by the models: a custom GPU optimization, a high-resolution Venus map assembled from existing photos, and one more Anthropic hasn’t detailed. That’s the pattern now — ship the model, then let the outputs make the case. Whether zero data retention moves the needle for enterprise adoption remains to be seen, but it removes one of the biggest objections CIOs have had about Anthropic’s flagship line.
