The United States may have imposed export restrictions on one of the most advanced AI models in the world because of fears it had already slipped into the wrong hands. According to a new report from Semafor, the White House's decision to limit exports of Anthropic's Mythos model was driven in part by concerns that a group linked to China had gained access to it.
What the report claims
The central worry, according to the report, is that if the Chinese government actually had access to Mythos 5 or Fable 5, it would represent a serious national security risk. These are frontier AI models — the kind capable of advanced reasoning, coding, and analysis — and Washington has spent the past few years trying to keep the most powerful systems out of the reach of strategic rivals.
The concern goes beyond simply using the model. Officials reportedly fear the Chinese government could attempt to reverse engineer it through a method known as distillation. In distillation, a "student" AI is trained on the outputs of a more advanced "teacher" model, learning to replicate its behaviour without needing the original underlying technology. In effect, a rival could clone much of a cutting-edge model's capability by studying how it responds.
Mixed signals from the White House
So far, the White House has not confirmed the report. A post on X by Trump adviser David Sacks did not mention China at all. Instead, Sacks focused on a different rationale for the restrictions, leaving the national security angle unconfirmed publicly.
That gap between the reported reasoning and the official messaging is notable. Export controls on AI have become one of the most sensitive levers in the US-China technology rivalry, and governments are often careful about how much they reveal regarding intelligence-driven decisions. The lack of a clear public explanation has left observers piecing together the motivations from competing accounts.
Why distillation worries officials
The fear of distillation is not hypothetical. As AI labs race to build ever more capable models, the question of how to protect that investment — and the strategic advantage it confers — has moved to the centre of policy debates. A model that took enormous computing power and research to build could, in theory, have its behaviour partially copied at a fraction of the cost.
For the US, that raises the stakes of any unauthorized access. It is not only about a competitor using a tool, but about a competitor potentially learning enough to build something comparable on its own.
The bigger picture
The episode underscores how AI has become a flashpoint in international security, not just a commercial technology. Export restrictions, once reserved largely for weapons and sensitive hardware, are increasingly being applied to software models themselves. As frontier AI systems grow more powerful, expect more decisions like this one — made quietly, justified vaguely, and shaped heavily by concerns about who else might be watching.
For now, the key claims remain unconfirmed, and Anthropic, the White House, and the alleged China-linked group have offered little clarity.
Source: The Verge, reporting on a Semafor story.
