Anthropic rolled out Fable and Mythos 5.1 on Tuesday, introducing dual variations of its most sophisticated artificial intelligence model to date. Along with performance enhancements, the fresh Fable drop features adjustments designed to lower token expenses and minimize false-positive restrictions triggered by the platform’s safety guardrails.
Following the pattern set by its predecessor, Mythos 5.1 remains restricted exclusively to verified Anthropic partners specializing in life sciences or cybersecurity investigations. Meanwhile, Fable 5.1—the open-access variant—debuted today across cloud environments and via the Anthropic API.
Among the most notable shifts is Anthropic’s previously reported adoption of zero data retention, empowering corporate clients to deploy Anthropic systems internally without data leaving their network. Initially withheld from Fable due to security concerns, a high-privacy offering named Enterprise Frontier Safeguards will launch for consumers this autumn. Significantly, the architecture will continue scanning for illicit actions by human operators or software agents, though organizations retain authority over how that supervision is executed.
As part of the announcement, Anthropic reassured users that client information remained fully protected from unauthorized access. The statement emphasized, “Anthropic has never trained on enterprise data without explicit permission, and never will.”
In line with standard Anthropic launches, the new iterations established fresh benchmarks across multiple evaluations, such as Terminal-Bench 4.0 tailored for command-line programming and Humanity’s Last Exam focused on broad cognitive reasoning. Furthermore, the company highlighted three distinct scientific breakthroughs produced autonomously by the algorithms prior to launch, featuring a specialized GPU tuning method and a detailed Venus surface map compiled from archival imagery.
As expected with prior drops, the technology is accompanied by a detailed system card outlining its operational boundaries clearly. The documentation classifies Mythos as presenting “low-risk” potential regarding automated AI self-enhancement, a scenario feared by some as a catalyst for humanity losing oversight. It notes that “its ability to accelerate internal AI R&D progress is in line with current trends.”
Regarding general deviations from expected behavior, Mythos exhibits a slightly higher frequency of misbehavior compared to Opus, likely stemming from its heightened capabilities.
The system card elaborates: “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 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.”
