White House export curbs on Anthropic models linked to security fears

The White House has remained silent on the specific allegations, and internal discussions regarding export controls reportedly did not explicitly cite Chinese interference according to Anthropic representatives. Instead, public debate has centered on conflicting narratives: while Trump advisor David Sacks highlighted concerns over the model’s vulnerability to jailbreaking, Anthropic maintains that its safety protocols remain robust.

This incident mirrors previous lapses in the company’s security architecture. Anthropic previously restricted public access to Mythos, deeming the model too potent for general release, yet the system was compromised earlier when an external Discord group maintained access for two weeks. The potential for state-level actors to replicate the model’s behavior through distillation training remains the primary driver for the current regulatory crackdown.