Tap Off

On Thursday, June 12, at 5:21 PM Eastern Time, Anthropic received a letter from the US government. The directive, citing national security authorities, ordered the company to suspend all access to its Fable 5 and Mythos 5 AI models by any foreign national, whether inside or outside the United States. The net effect was total: Anthropic had to disable Fable 5 and Mythos 5 for all customers worldwide to ensure compliance. Access to all other Anthropic models remains unaffected.[1]

The stated reason was a potential jailbreak. According to Anthropic's own statement, the government provided only verbal evidence of a narrow, non-universal jailbreak, essentially asking the model to read a specific codebase and fix software flaws. Anthropic reviewed the report and concluded that the capability level displayed was "widely available from other models," including OpenAI's GPT-5.5, which offers similar cybersecurity features publicly. No harmful result from the jailbreak was disclosed. The vulnerabilities it found were minor and already known.[2]

The implication, though, was unmistakable. The US government demonstrated that it can, with a single letter, turn off access to frontier AI for the entire non-American world. The "tap" exists. It was used.

France's response was immediate. On Tuesday, Prime Minister Sebastien Lecornu announced that the Direction Generale de la Securite Interieure (DGSI), France's domestic intelligence agency, will end its contract with Palantir, the American data analytics giant co-founded by Peter Thiel. "We cannot accept new strategic dependencies in the digital sphere," Lecornu said in a video statement. France will invest 655 million euros ($760 million) in developing its own AI capabilities instead.[3]

The timing was deliberate. Palantir, which has worked with the US government to identify targets and undocumented immigrants, and which is deeply embedded in European security infrastructure, suddenly looked like a liability. The same government that turned off Anthropic's models could, in theory, do the same with Palantir's data platform. Britain's parliament had already raised concerns. A report from the Science, Innovation and Technology Committee warned that reliance on a small number of US-based providers "represents a clear vulnerability" that could leave public services "at the mercy of foreign actors." London's mayor blocked the Metropolitan Police's bid to work with Palantir. The NHS contract is under review.[4]

Anthropic, to its credit, was transparent about what happened. Its statement detailed the government's directive, the nature of the jailbreak concern, and its own assessment that Fable's safeguards remain substantially more effective than any previous model's. It pointed out that no universal jailbreak has been found, that perfect jailbreak resistance appears impossible for any model provider today, and that the capability the government described is already available from competing products with no access restrictions. The company also noted its defense-in-depth strategy, including mandatory 30-day data retention for Fable users, designed to detect and shut down attacks quickly.[5]

But transparency about the mechanism does not change the outcome. A foreign government decided, without public process or disclosed evidence of actual harm, that non-Americans could not be trusted with a specific AI model. The model was turned off. Compliance was immediate. There was no appeal, no negotiation, no transition period. The tap was turned, and the water stopped.

For European policymakers, this is the moment the abstract debate about digital sovereignty became concrete. It is one thing to discuss supply chain risk in conference rooms. It is another to watch a frontier model go dark across an entire continent because a single government, in a single letter, decided it should. The Palantir contract cancellation is the first visible consequence. It will not be the last. Lecornu's 655 million euros is a down payment on a very long road.