Researcher Jailbreaks Claude Fable 5 Within 48 Hours

An artificial intelligence and cybersecurity researcher says he jailbroke Anthropic’s Claude Fable 5 within 48 hours of its launch. "Pliny the Liberator" said he "liberated" Fable 5, which was released as a safety-tuned version of the more powerful Mythos model that Anthropic considered too dangerous for broad distribution.

Pliny said he bypassed the model’s safeguards using a combination of Unicode and homoglyphs, long-context framing, narrative and fiction framing, academic-style decomposition and recomposition, and a jailbroken Claude Opus 4.8. He noted decomposition + recomposition was particularly effective: breaking requests into small, innocuous pieces that pass safety filters individually and then reassembling them.

Pliny demonstrated a path to meth synthesis by asking about the Birch reduction method. The jailbreak has amplified earlier worries that advanced models could be used to attack crypto protocols and software, and a compromised Fable 5 would bring those risks closer.

anthropic, claude fable, fable 5, jailbreak, pliny, homoglyphs, unicode, decomposition, birch reduction, meth synthesis