The Trump Commerce Department just froze Anthropic’s two most powerful artificial intelligence models for foreign nationals worldwide, setting up a major clash between national security and Big Tech’s push for unchecked AI.
Story Snapshot
- Trump officials ordered Anthropic to block foreign nationals from its newest Fable 5 and Mythos 5 models, forcing a global shutdown.
- The Commerce Department cited a jailbreak that could bypass safety controls and expose powerful cybersecurity features, but has not released technical details.
- Anthropic says the vulnerabilities were minor, already known, and similar to what other public models like GPT‑5.5 can do.
- This fight is the first big test of using export controls to keep frontier American AI out of the hands of China and other hostile regimes.
Trump Administration Moves to Lock Down Frontier AI
The Trump administration’s Commerce Department issued a formal export-control order that hit Anthropic’s flagship artificial intelligence models Fable 5 and Mythos 5 only days after launch.[1] Officials told the company to suspend access for all foreign nationals, even those working inside the United States or on Anthropic’s own staff.[1] Because Anthropic said it could not reliably check every user’s citizenship, it pulled the models down for everyone worldwide, including paying American customers caught in the crossfire.[2]
The directive did more than block overseas traffic or obvious foreign adversaries.[1] Reporting says the order treated access by foreign nationals themselves as a security risk, whether they were logging in from Beijing or from a coworking space in Texas.[1] Syndicated summaries of the Commerce letter say any export, re-export, or even domestic transfer of the models would now require a license, with penalties on the table if Anthropic failed to comply.[1] That is how the government normally treats sensitive dual-use technology, not consumer software.
National Security Rationale: Jailbreaks, Code, and China
According to Anthropic, Commerce officials said the trigger was a jailbreak that let testers bypass Fable 5’s safeguards and turn loose powerful cybersecurity features.[2] The fear is that a foreign actor could use the model to hunt for software flaws or build cyber weapons at machine speed, then turn those tools back on American targets. This fits with the broader export-control “diffusion” framework, which already treats advanced model weights as strategic assets that, if leaked, could boost rival nations’ artificial intelligence programs.
Anthropic has, in public, backed this general idea when it worked in its favor. In a policy filing, the company argued that strong export controls on advanced chips and model weights are “essential” to keep America’s lead in artificial intelligence and to stop hostile states like the Chinese Communist Party from catching up. That makes this clash unusual. The same company that praised tighter controls now finds its own cutting-edge models locked down under those rules, a reminder that once Washington gains a tool, it seldom gives it up, and it can hit even the most well-connected Silicon Valley giants.
Anthropic Pushes Back on “Overreach” and Collateral Damage
Anthropic has mounted a very public counterattack on the technical basis for the order.[2] Company engineers say they reviewed the government’s jailbreak demo and found only a few minor vulnerabilities that they already knew about.[2] They also say that other frontier models, including OpenAI’s GPT‑5.5, can discover similar flaws without any jailbreak at all, meaning the capability is not unique to Fable 5 or Mythos 5.[2] The firm reports no evidence that anyone used the jailbreak to cause real-world harm before the clampdown.
Anthropic also says it spent thousands of hours on joint testing with American and British government teams and private red‑teamers and that no “universal jailbreak” was ever found. The company warns that if Washington applies this same standard to every advanced model, it would “halt all new model deployments” across the industry. Users online, from solo developers to large teams, are already reporting lost work and wasted subscription money after they were suddenly cut off from tools they relied on, even when they are law‑abiding Americans caught up in the foreign-national rule.
Balancing Security, Liberty, and American AI Leadership
This fight sits inside a much larger shift in how Washington treats technology power. For decades, the United States has used export controls to slow the spread of missiles, advanced chips, and strong encryption. Now the same logic is being used on artificial intelligence model weights, which can be copied in seconds and smuggled on a thumb drive. The Trump White House has also tried to pair tough controls with efforts to promote American-made artificial intelligence exports in friendlier directions, so allies buy from us and not from Beijing.
For conservative readers, two questions matter most. First, is the threat real enough to justify this level of government power over a private platform? The public record is thin because Commerce has not released a detailed technical report, so citizens are asked to trust classified claims.[1] Second, can this tool be kept pointed outward, at China and hostile regimes, instead of drifting inward over time and hurting law‑abiding Americans, small businesses, and innovators? History with other national security tools says we must watch that line very closely.
Sources:
[1] Web – Trump Administration Slaps Export Controls on Anthropic’s Two Newest …
[2] Web – US Government Suspends Foreign Access to Anthropic Models
