Anthropic restricted Claude Fable 5 for Chinese labs — and faced criticism from the West

Anthropic released Claude Fable 5 — a public version of its most powerful Mythos model, which it had held back since April. The new model launched with built-in restrictions specifically designed to prevent Chinese AI labs from accessing its most powerful features. The loudest complaints came not from China, but from the West.

Fable 5 instead of Mythos

On June 9, Anthropic officially opened public access to Claude Fable 5. This is a "softened" version of the flagship Mythos model — an original development that the company had withheld since April. The reason for the delay was not publicly disclosed, but based on available information, it was precisely the outstanding capabilities of Mythos that raised serious safety concerns for Anthropic. Now the logic has become clearer.

Anthropric operates on a two-tier access scheme: the full Mythos remains closed — it is available for internal research and a limited circle of verified partners with separate agreements. Fable 5 is what the company is willing to offer to the wider audience: a model with the same architectural foundations, but deliberately calibrated in capabilities and having undergone additional restrictive filters. A similar scheme has long been used in the industry: major AI companies often maintain distance between internal and public versions of their flagship models. The difference is that Anthropic is one of the first to openly acknowledge that geopolitics is part of the logic behind these decisions.

Barrier against Chinese labs

The central feature of Fable 5 is built-in measures designed to prevent Chinese AI labs from accessing the model's most powerful capabilities. The company does not disclose the exact technical composition of these restrictions, but their existence is an officially confirmed fact. This fits into a broader trend. American technological policy, starting with restrictions on chip exports and high-performance computing systems, is consistently shifting the logic of control to the level of software and AI models. Some companies are moving in this direction voluntarily and preventatively, without waiting for regulators to make it mandatory.

Criticism from an unexpected quarter

The paradox of the release is that the sharpest complaints came not from Chinese competitors — whom the measures are designed against — but from Western developers, precisely the audience for which Fable 5 is officially intended. The essence of the complaints boils down to several key points:

• restrictions are opaque — there are no clear public criteria for what exactly is being blocked and why
• legitimate research and commercial tasks unrelated to any real risks fall under automatic filters
• there is no clear appeal mechanism or way to justify the good faith of a specific project
• legal and operational uncertainty creates difficulties for companies building products on top of the API

This is a classic problem with any control system: measures aimed at potential offenders inevitably create friction for good-faith users. The more opaque the mechanism — the stronger the irritation among those who don't understand why their request was rejected.

What this means

Anthropric is consistently building the concept of a "secure frontier": the most powerful capabilities are available only to those the company trusts directly, while the public API is an intentionally limited version taking into account geopolitical risks. For the market, this is a clear signal: the era of completely open top-tier models without built-in regulatory barriers is ending. Developers will have to build products in conditions where access rules are determined not only by the technical characteristics of the model, but also by global politics.