Hajra Naz
On Tuesday, Anthropic released Fable and Mythos 5.1. These represent twinned versions of the company’s most advanced AI model. Beyond general performance upgrades, the new Fable release includes specific improvements. That includes changes meant to reduce token cost. It also includes reductions in false-positive restrictions from the model’s safeguards.
Similar to the previous version, Mythos 5.1 remains limited in availability. It will only be available to registered Anthropic partners. Specifically, this applies to partners engaged in cybersecurity or life sciences research. Fable 5.1, the unrestricted version, is different. It’s available starting today. Users can access it on cloud platforms. It’s also accessible through the Anthropic API.
One of the most significant changes involves data handling. Anthropic has embraced zero data retention, as previously reported. This allows clients to run Anthropic models on their own infrastructure. Notably, this happens without any data outflows to Anthropic itself. Previously, this option wasn’t available for Fable, due to security concerns.
Read More: Anthropic Explains Why Claude Fable 5 Safety Controls Were Not Visible
Now, a new high-privacy service will roll out this fall. It’s called Enterprise Frontier Safeguards. Importantly, the system will still monitor for misuse. This applies to both AI agents and human users. However, clients will control exactly how that monitoring takes place.
As part of this announcement, Anthropic offered direct reassurance to customers. The company confirmed that customer data hadn’t been inappropriately accessed. “Anthropic has never trained on enterprise data without explicit permission, and never will,” the announcement reads.
As is typical with Anthropic releases, these new models set several records. That includes strong results on Terminal-Bench 4.0, which measures CLI-based coding performance. It also includes strong results on Humanity’s Last Exam, a benchmark focused on general reasoning.
Anthropic also shared something notable alongside this release. The company released three novel scientific findings. These were generated by the models before their public release. That includes a custom GPU optimization. It also includes a high-resolution map of Venus, assembled entirely from existing photographs.
Consistent with previous releases, these models come with a detailed system card. This document explains the models’ capabilities in fairly straightforward terms. According to the system card, Mythos received a “low-risk” rating. This rating specifically addresses concerns around automated AI development.
Read More: Anthropic rolls out Claude Mythos 5 and Fable 5 for everyone
That refers to scenarios where AI improves itself. Some view this as a potential trigger for loss of human control. The system card states that Mythos’s ability to accelerate internal AI research and development remains in line with current trends.
In terms of general misbehavior, there’s a notable difference between models. Mythos is slightly more prone to misbehavior compared to Opus. This may result from its enhanced overall capabilities.
“Mythos 5.1 is a slight regression in overall misaligned behavior compared to Opus 5 and an improvement over Mythos 5 and Claude Sonnet 5,” the system card reads. ” 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.”
