Trade “castration” for “unblocking”? Fable5 returns tomorrow, but it is no longer the original Fable 5.

After weeks of restrictions, Anthropic's Fable 5 has finally been allowed back online, but at the cost of significantly tightened core capabilities.

On Tuesday, June 30 local time, the U.S. Department of Commerce officially lifted export restrictions on Anthropic's Fable 5 and Mythos 5 models. Anthropic subsequently confirmed on X platform that access will be restored on Wednesday.

However, Anthropic posted on X platform, stating that compared to previous versions, the relaunched Fable 5 has added stricter safety classifiers. Some requests for programming, debugging, etc., will be automatically downgraded to the older generation model Opus 4.8, meaning the scenarios in which users can truly invoke Fable 5's capabilities have been significantly compressed.

This change quickly sparked heated discussion among developers. Many industry insiders believe that although Fable 5 has been nominally reopened, its most competitive capabilities have been "neutered," making it more like a heavily restricted version rather than the full model previously released.

Fable 5 is back, but core capabilities have been "downgraded"

The company announced a major adjustment: the new version of Fable 5 has added a set of safety classifiers to identify requests that may involve sensitive cybersecurity capabilities. Anthropic acknowledged that the cost of this classifier is that it will "more frequently misjudge normal requests," including routine programming, code debugging, and other tasks.

The official statement reads:

The new classifier also means that in daily coding and debugging tasks, more normal requests will be incorrectly identified.

For these identified requests, the system will not invoke Fable 5 but will automatically switch to the previous generation Opus 4.8 model to complete them. Developers generally believe that this means Fable 5's strongest code generation capability is actually no longer fully usable.

Developer Wise stated: Fable 5 is no longer Fable 5. To exchange for the lifting of export restrictions, Anthropic added new safety classifiers while reassigning a large number of routine programming and debugging tasks to Opus 4.8.

Well-known crypto market analyst Miles Deutscher also pointed out that the new version of Fable 5 not only restricts coding capabilities but also allows users to use the model only within approximately 50% of their quota, with the remaining requests still falling back to the old model. Therefore, this "relaunch" is more like a restricted opening rather than a true restoration of original capabilities.

From full blockade to conditional unblocking

This adjustment is the result of weeks of negotiations between the U.S. government and Anthropic.

On June 12, the U.S. Department of Commerce, citing national cybersecurity risks, banned Anthropic from providing users with the two cutting-edge models, Fable and Mythos. The biggest concern for regulators centered on the so-called "jailbreak" risk, i.e., that users might bypass the model's safety restrictions to gain more powerful cyberattack capabilities.

Subsequently, the U.S. government last week first allowed limited access to Mythos 5 for certain U.S. companies and government agencies, while the consumer-grade product Fable 5 remained blocked. According to previous media reports, to promote the unblocking, Anthropic ultimately adjusted its negotiation strategy, no longer insisting on the position that "the risk of jailbreaking cannot be completely eliminated," but instead promising to add stricter safety protection mechanisms to reduce the probability of risk occurrence.

The safety classifier added to the now-relaunched Fable 5 is the main outcome of the compromise between the two sides.

AI regulation enters the stage of "capabilities available, but must be controllable"

The restoration of Fable 5 also further reflects the changing approach to AI regulation in the U.S.

Compared to directly banning model access, the U.S. government prefers to require companies to implement ongoing supervision of cutting-edge models by adding safety classifiers, restricting sensitive capabilities, dynamically downgrading models, and so on.

Anthropic is not the only company affected. Previously, OpenAI's latest model ChatGPT-5.6 was also only opened to a small group of users under the requirements of the White House and relevant regulatory agencies, with a significantly slowed release pace. OpenAI management has publicly stated that they hope to establish a clear and stable AI model evaluation and approval mechanism as soon as possible to reduce the uncertainty brought by temporary regulation.

For the entire AI industry, the "unblocking" of Fable 5 may signal a new regulatory paradigm taking shape: Cutting-edge models can be released, but the most sensitive and powerful capabilities will increasingly be dynamically controlled through safety classifiers, capability downgrades, and access restrictions. Developers get a usable Fable 5, but it may not necessarily be the original complete Fable 5.

Risk Warning and Disclaimer

        Market risk exists, and investment requires caution. This article does not constitute personal investment advice, nor does it consider the specific investment objectives, financial situations, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investment based on this is at your own risk.