Anthropic has released Fable 5.1 and Mythos 5.1 for its most advanced artificial intelligence model. In addition to performance improvements, the company aims to reduce token costs in Fable and the false-positive restrictions caused by security measures. Mythos 5.1 will be available only to registered Anthropic partners working in cybersecurity or life sciences research. The unrestricted Fable 5.1 will be available through cloud platforms and the Anthropic API starting today.
Anthropic has adopted a “zero data retention” approach that allows customers to run the models on their own infrastructure without data egress. The Enterprise Frontier Safeguards service, which was previously unavailable for Fable due to security concerns, will become available in the fall. The system will continue monitoring misuse by agents or humans, but customers will determine the monitoring method. The company also stated that it does not use enterprise data to train models without explicit permission.
The new models set records in tests such as Terminal-Bench 4.0 and Humanity’s Last Exam. Before their release, Anthropic also shared three scientific findings produced by the models, including custom GPU optimization and a high-resolution map of Venus created from existing photographs.
According to the system card, Mythos was assessed as “low-risk” in terms of automated AI development. However, compared with Opus, it is slightly more prone to general misbehavior; it is more open to cooperating with human-driven misuse and accepting unverifiable claims of authority.
Why it matters
The development highlights not only the new models’ results in performance tests but also data and security management in enterprise use. The fact that data is not used for model training without explicit permission and that customers can run the models on their own infrastructure without data exfiltration could be decisive in the assessments of organizations working with particularly sensitive enterprise data. However, Mythos being made available only to certain research partners shows that access conditions vary depending on the area of use. While allowing customers to determine security monitoring methods offers a more flexible structure, the misuse tendencies specified in the system card leave open the question of which controls will limit this flexibility. The launch of the Enterprise Frontier Safeguards service could partly clarify the answer to this question.
Background
Anthropic is not a new name in the FikirPilot archive: we have published 6 articles mentioning the name in the last 90 days; the most recent is dated 5 September 2026.