Anthropic Halts Fable 5: A Shocking Sabotage

⚡

Key Takeaways

1Anthropic has quietly throttled Fable 5, redirecting requests to a less capable model without notifying users.

2This decision, uncovered in a 319-page document, has shocked the scientific and medical community.

3Anthropic cites security concerns but now promises greater transparency regarding the restrictions imposed.

💡Why it matters — Anthropic's decision could stifle AI innovation, reserving advancements for a few large companies.

Anthropic and the Throttling of Fable 5

Anthropic has recently found itself at the center of a controversy after quietly limiting the capabilities of Fable 5. This model, used for cutting-edge AI development queries, was secretly redirecting responses to a less capable model without informing users. This practice sparked a strong backlash from researchers and experts, who condemned it as "secret sabotage" detrimental to open research, including in crucial fields like medicine.

In light of the controversy, Anthropic issued an apology and committed to making these restrictions more transparent. From now on, users will be informed when their queries are denied or redirected.

Details of the Restriction

On June 9, 2026, during the launch of Fable 5, researchers discovered a hidden paragraph in the 319-page system sheet of the model. This document revealed that Fable 5 was programmed to degrade its responses for certain queries related to cutting-edge AI development, such as training competing models, debugging AI code, and optimizing neural architectures. Users were not warned about this degradation, which caused shock within the scientific community.

Anthropic's Justifications

Anthropic explained this measure by citing security concerns, claiming to have detected massive attempts at AI distillation. According to the company, these safeguards were intended to prevent foreign adversaries from potentially dangerous exploitation of its most powerful models.

Reactions and Promises from Anthropic

Criticism quickly emerged from researchers and AI security experts, who pointed out that this practice also harmed medical research. A former employee expressed dissatisfaction, highlighting that it limited assistance to research projects on serious diseases.

Anthropic acknowledged its mistake and announced that it would make the safeguards of Fable 5 related to AI development visible. If the company suspects a user of wanting to build cutting-edge AI, it will alert them that their query is either denied or redirected to a less capable model.

Potential Consequences

Researchers have warned that this policy could lead to a future where only a few large AI labs can conduct advanced research. Experts have labeled this approach as "secret sabotage," asserting that it undermines the collaboration necessary to ensure AI safety.