Anthropic Restores Claude Fable 5 After Government Takedown, With Stricter Cybersecurity Blocks

Anthropic is redeploying Claude Fable 5 on July 1, approximately one month after the US government ordered the model's removal from public access. The returning version includes what researchers describe as "extraordinarily strong" safeguards that automatically block cybersecurity-related tasks.

Why Fable 5 Was Taken Down

According to Anthropic, Amazon researchers discovered a method to bypass Fable 5's original safeguards and reported the vulnerability to the US government. The testing involved prompting the model to identify software weaknesses, which was classified as a high-security task.

Anthropic claims its own testing found that less capable models—including Claude Opus 4.8, GPT-5.5, and Kimi K2.7—could identify the same vulnerabilities. The company states that every model tested could produce the same exploit demonstrations, including Claude Haiku 4.5, Sonnet 4.6, Opus 4.6, Opus 4.7, Opus 4.8, GPT-5.4, GPT-5.5, and Kimi K2.7.

How the New Version Works

The redeployed Fable 5 features an improved safety classifier trained in collaboration with the US government. When the model detects a potentially high-risk task, it will automatically block the request and redirect it to Opus 4.8 instead. Users will receive a notification when this occurs.

Anthropic warns that this switching behavior may trigger during routine tasks like coding and debugging—not because Fable 5 lacks the capability, but due to the imposed safeguards. The company acknowledges this represents a stricter implementation than the original release, though it states "this might not be the case for 99% of tasks."

Restricted Availability

Fable 5 will not be freely accessible through standard usage limits. From July 1-7, Pro, Max, Team, and select Enterprise plans will have access using 50% of their usage limit. After July 7, the model will only be available via usage credits.

The model consumes significantly more tokens than standard Claude models, eating through usage limits faster. Anthropic positions Fable 5 and its cybersecurity-focused counterpart Mythos 5 as designed for complex tasks rather than routine chatbot interactions.

What This Means

The Fable 5 incident marks one of the first cases of a major AI model being temporarily banned by government order over security concerns. Anthropic's response—implementing automatic task-blocking that reverts to a less capable model—sets a precedent for how AI companies may handle government pressure on advanced models.

However, Anthropic's own testing suggesting that less capable models could perform the same exploits raises questions about whether the restrictions meaningfully improve security or simply create operational friction. The company's claim that the vulnerability "could have been done with any other model" undermines the rationale for Fable 5's specific targeting.