The US Just Shut Down Anthropic's Most Advanced AI Models

The US government has ordered Anthropic to suspend foreign access to Fable 5 and Mythos 5, citing national security. Anthropic calls it a misunderstanding over a narrow jailbreak. The deeper story is twofold: frontier models are becoming strategic national assets and the people we have placed in charge of them still behave like rivals who won't hold a hand on stage.

The United States has just crossed a new line in the politics of artificial intelligence. Anthropic says the US government, citing national security authorities, ordered it to suspend access to Fable 5 and Mythos 5 , its two most capable models for any foreign national, whether inside or outside the United States. The directive reportedly reaches foreign nationals working inside Anthropic itself. Unable to separate users by citizenship in real time, the company says it has no choice but to disable both models for everyone. Access to its other models is unaffected. This is not a normal product outage. It is one of the clearest signs yet that frontier AI is being pulled into the same geopolitical logic that already governs chips, satellites, cryptography and weapons. According to Anthropic, the government's concern centers on a method for bypassing "jailbreaking" Fable 5. The company says it reviewed a demonstration of the technique and found that it surfaced only a small number of previously known, minor vulnerabilities. By its account, the jailbreak amounts to little more than asking the model to read a codebase and fix its flaws, and the capability is widely available elsewhere, including in OpenAI's GPT-5.5. It is the kind of thing security defenders do every day. So Anthropic is not saying the government invented the concern. It is saying the response is wildly out of proportion. Applied across the industry, the company warns, this standard "would essentially halt all new model deployments for all frontier model providers." That distinction matters. Every frontier lab now lives inside the same paradox: the more powerful a model becomes, the more governments treat it as strategic infrastructure. A system that can write code, reason across systems, automate research and probe for weaknesses is also, in the wrong hands, a weapon. The same capability that helps a defender patch a system helps an attacker find the hole. The same intelligence that accelerates discovery accelerates misuse. This is why AI safety is no longer only a question of corporate ethics. It is becoming a question of national security. For years, Washington aimed its AI strategy at hardware GPUs, fabrication equipment, the semiconductor supply chain. The logic was clean: if compute is the engine, controlling the engine controls who can build. But this directive points to a second phase. Governments may now try to control not just the machines that create frontier AI, but the models themselves. That is a far messier world. Chips are objects. They can be tracked, licensed, shipped and seized. Models are software deployed across clouds, APIs, enterprise tools and internal workflows, used by hundreds of millions of people. Restricting them by nationality is technically, legally and operationally close to impossible. So the state reaches for the only lever that actually works: the off switch. To keep two models from a few, Anthropic is pulling them from all. This sets a precedent that will outlive the incident. If one narrow jailbreak claim can trigger a government-ordered shutdown of a commercial model, future releases may quietly require national-security clearance in practice long before any transparent law says they must. Anthropic itself argues the state should be able to block unsafe deployments, but only through a process that is transparent, fair, and grounded in technical fact. This, it says, was none of those things. And here is where the story stops being about models and starts being about the people who control them. Because the uncomfortable question underneath all of this is not what these systems can do. It is who we have put in charge of them. Step back four months. February 2026, the India AI Impact Summit. Narendra Modi gathers the most powerful people in technology onto one stage and reaches for the oldest gesture of unity there is hands joined and lifted together for the cameras. Sundar Pichai does it. Meta's Alexandr Wang does it. And then there are the two men standing side by side: Sam Altman of OpenAI and Dario Amodei of Anthropic. They do not clasp hands. They do not look at each other. Each raises a fist, alone. Altman later said he simply wasn't sure what was happening. Perhaps. But the clip traveled the world for a reason. It was instantly legible. Two of the people racing to build the most consequential technology in human history could not perform the smallest act of public grace to hold a rival's hand for three seconds in front of a prime minister. This is the part few want to say plainly. The people steering the trajectory of artificial intelligence have not, by any visible evidence, shown the moral and social maturity the role demands. They feud through Super Bowl attack ads. They freeze each other out on stage. Amodei left OpenAI years ago, in part because he believed it had drifted too far toward commercial ambition; the rivalry has only hardened since. These are not the manners of statesmen entrusted with a civilizational technology. They are the manners of competitors who cannot get past each other. And yet we are asked to trust their judgment about the most serious questions imaginable: catastrophic risk, who may and may not touch a frontier model, what counts as too dangerous to release. Given a stage on which to model maturity for the entire world, two of them chose pettiness. That should unsettle us more than any jailbreak. Because the power is now real, and it is now political. Governments are treating these models as strategic assets and they are right to. But strategy assumes a steady hand on the asset. What the India stage revealed, and what the rivalry keeps revealing, is that the hands are not steady. They are competitive, wounded, and human in the smallest sense of the word. Fable 5 and Mythos 5 may come back. Anthropic believes the shutdown is a misunderstanding and says it is working to restore access. But even if the models return tomorrow, two facts remain. Frontier AI is no longer just a product category. It is a controlled strategic asset. And the people holding it have not yet shown they are ready to. The AI race has entered its export-control era. Intelligence is now a matter of national security, and access will never again be just access. But the deeper border isn't the one the government is drawing this week