The US government issued a directive on June 12 ordering Anthropic to immediately suspend access to its most advanced AI models for all foreign nationals. That includes the company’s own international employees.
The order targets Anthropic’s Fable 5 and Mythos 5 models, which had been publicly rolled out just days earlier. A reported “jailbreak” vulnerability triggered national security alarms, with concerns that the models could be exploited to identify software vulnerabilities. Anthropic complied swiftly, disabling global access to both systems.
Here’s the thing: this is the first time US export controls have been used to compel a major AI lab to shut down its frontier models. Not a fine. Not a warning letter. A full access revocation, delivered at approximately 5:21 p.m. ET on a Thursday.
What happened and why it matters
The directive came from the Trump administration as part of a broader strategy treating advanced AI technologies as national security assets.
Anthropic’s response was notably conflicted. The company complied immediately, reasoning that implementing targeted restrictions, like blocking specific countries or user groups, would be infeasible on short notice. But it also publicly criticized the government’s approach.
Anthropic argued that a minor potential vulnerability should not justify measures that effectively halt international access to commercially deployed models. The company called for a fair and transparent governmental process for imposing such restrictions, rather than the ad hoc directive it received.
Over a dozen cybersecurity experts have since criticized the ban, calling it a “dangerous” move.
The broader context of AI export controls
The jailbreak vulnerability at the center of this controversy reportedly raised concerns about the models being used to discover exploitable weaknesses in software systems. Jailbreaks in AI refer to techniques that bypass a model’s safety guardrails, tricking it into producing outputs it was designed to refuse. In this case, the concern was that a jailbroken version of Fable 5 or Mythos 5 could serve as a tool for offensive cyber operations.
What this means for investors and the AI industry
For investors in the AI sector, this episode introduces a new category of risk. The possibility that a commercially deployed AI product can be shut down internationally, with no prior notice, on the basis of a security concern changes the calculus for anyone funding frontier AI development.
Anthropic’s international customers, enterprise clients who had integrated Fable 5 and Mythos 5 into their workflows, suddenly lost access.
The precedent matters more than the specific incident. If the government can do this to Anthropic, it can do it to OpenAI, Google DeepMind, or any other lab pushing the frontier.