The US government has directed Anthropic to restrict access to its most powerful AI models to US nationals only, marking an unusual move by the White House into AI safety oversight.

What happened: On Friday at 5:21 pm, Anthropic received a directive from the US government instructing it to limit access to its most powerful AI models—including Fable 5 and Mythos—to US nationals only. The government claimed to have found a 'jailbreak' (a way to bypass some of Fable 5's safety guardrails), though Anthropic disputes the severity of the issue. Anthropic pushed back in a blog post, disagreeing with the government's decision.

Why it matters: AI safety experts have long called for a 'pause' on developing the most powerful AI models, and Anthropic itself made such a call in a recent blog post. This directive may represent the first concrete action on that concern, though it comes from an unexpected source: the White House rather than industry consensus. If enforced across the sector, the restriction could make it economically unviable or illegal to develop frontier AI models, since many foreign nationals work at major AI labs.

What to watch: The White House has been taking a more aggressive approach to AI regulation, spearheaded in part by National Cyber Director Sean Cairncross. A recent executive order directs agencies to create a voluntary framework under which AI developers could give the government secure access to certain powerful frontier models for up to 30 days before a broader release. Whether this directive becomes actual policy affecting other AI labs remains unclear.