Anthropic intros Fable and Mythos 5.1 models with a safety wall

Anthropic announced the release of two new models, Claude Fable 5.1 and Claude Mythos 5.1. Technically, it is one and the same model at the peak of its capabilities, but with two different levels of safeguards.

The end of a unified interface for everyone

Fable 5.1 is the publicly available version intended for general corporate and research use, where the capabilities to generate code for cyberattacks or life sciences designs are heavily restricted. Mythos 5.1, on the other hand, offers these risky capabilities unlocked, but is available exclusively through trusted access programs, meaning harder verification of who uses the model and for what.

Distributing performance based on trust

This move confirms the trend outlined by the European AI Act. Developers are stopping the practice of packing all capabilities into one public model with universal prompt alignment. Instead, they release a stripped-down version for the public and hide full power behind B2B verification gates.

Problems with certification transparency

While the principle of separation makes sense, the implementation raises questions about who actually decides on granting access to Mythos 5.1. There is no independent certification body, and Anthropic retains the role of sole judge over which security researcher is trustworthy enough.

Commercial impacts on red teaming

For security researchers and offensive security professionals, this means top models will no longer be available for just the price of a monthly subscription. The availability of dangerous capabilities is starting to resemble the market for explosive precursors, not software.

Lilith's verdict

The unified API for everyone is over. Models are becoming weapons for which you need not only a gun license, but also a vetted supplier who will lend them to you.