Three days after launching Fable 5 and Mythos 5 on June 9, Anthropic received a letter from Commerce Secretary Howard Lutnick: comply with a foreign-national access ban or face criminal and civil penalties. Anthropic complied on June 12, suspending access for foreign nationals among the hundreds of millions already on them.
Lutnick’s letter cited federal laws allowing export controls on civilian technology that could be used for intelligence purposes by an adversary’s military. The trigger: the government believed someone had found a method to jailbreak Fable 5. Anthropic’s statement called it a “narrow potential jailbreak” and said its layered monitoring and data retention systems were built precisely to catch and contain that kind of exploit.
The company’s sharpest argument isn’t about its own models. Anthropic said applying this standard across the industry “would essentially halt all new model deployments for all frontier model providers.” That’s a precise claim about precedent. If Commerce can invoke national security export controls over a single jailbreak report on a civilian AI model, every frontier lab from OpenAI to Google DeepMind is one exploit claim away from the same ultimatum. Lutnick’s original letter didn’t cite a specific violation, only federal laws that permit such controls.
Anthropic sent top members of its tech team to Washington over the weekend. As of June 16, representatives were meeting both online and at Commerce, with the White House also involved. Any AI company serving foreign nationals needs export counsel reviewing model capabilities before the next launch, not after the government letter arrives.
Nathan Zakhary