Pliny the Liberator Leaks Claude Fable 5 System Prompt on GitHub

An AI researcher named Pliny the Liberator has leaked the system prompt for Anthropic's Claude Fable 5, according to available records. The leaked information, posted on GitHub on June 10, 2026, provides extensive details about the model's personality, safety mechanisms, and behavior guidelines. The prompt revealed that Fable 5 shares the same underlying model as Mythos 5, with distinct safety filters.

It also disclosed a knowledge cutoff date of January 2026 and imposed strict copyright limitations on the model's outputs. Pliny claimed to have successfully jailbroken the model using advanced prompting techniques, but Anthropic disputed these claims, stating that the leaked document does not demonstrate a genuine jailbreak of Fable 5's safety systems.

The researcher's assertions about jailbreaking Fable 5's safety features were found to be unsupported by the provided examples, which consisted of general information available in public sources. As of the latest information, there is no confirmed leak of the system prompt for Fable 5.1, which was released three months later.