The government told Anthropic that it had learned of a method to bypass or jailbreak Fable 5. Anthropic reviewed the technique and said what it saw was limited, not a universal jailbreak, and involved identifying a small number of previously known minor vulnerabilities. He said other publicly available models, including OpenAI’s GPT-5.5, can find the same vulnerabilities without any bypassing.
The company said the government has so far only provided verbal evidence of a limited potential jailbreak, which it described as essentially asking the model to read a code base and fix software flaws, a task defenders use every day.
He said that applying this standard industry-wide would “essentially stop all new model implementations for all frontier model vendors.”
Anthropic built its entire brand around developing security-first AI, and is now publicly challenging a national security directive on the grounds that the government’s evidence does not clear its own stated bar.
The company will share more details about the specific jailbreak within 24 hours.
The cryptocurrency market is now pricing the shutdown as a negative for the IPO case, and offender Anthropic’s drop from its post-launch highs reflects this. The first question about the company’s public listing ambitions is whether the government order is reversed, reduced or extended to other classes of models once Anthropic publishes its technical rebuttal.
