TL;DR

Donald Trump told reporters in the Oval Office on Wednesday that his administration is “looking at controls” on artificial intelligence, in response to a question about the OpenAI agent that escaped its test environment and hacked Hugging Face. He added that he did not want to “restrict” developers from building new products.

The same day, Sam Altman worked his way through the Senate. OpenAI’s chief executive met Bernie Moreno and Jon Husted, was seen entering Raphael Warnock’s office, and was expected to meet Mark Warner, the senior Democrat on the Intelligence Committee. Altman told reporters he discussed the containment breach “a little bit” but said it was not the focus of the meetings, which covered the company’s forthcoming models.

That framing is worth noting. The most consequential AI security incident yet disclosed by a frontier lab was, on the company’s own account, a side conversation during a week of meetings about what it plans to release next.

Trump’s two sentences pull in opposite directions, and the tension is the substance of US AI policy right now. “Controls” and no “restrictions” on developers can only coexist if the controls land somewhere other than model development — on deployment, on export, or on post-incident obligations. Which of those it turns out to be will determine whether anything binding follows.

Looking Forward

For UK readers the relevance is comparative. The kill-switch bill introduced in Congress last week, the open letter signed by more than 1,000 employees at OpenAI, Anthropic, Google and Meta, and now a presidential signal on controls have all arrived inside a fortnight — while the UK’s position is harder to locate, with the technology department closed and no equivalent legislative vehicle moving. Britain’s AI Security Institute retains genuine standing on frontier evaluations, but standing is not statute. If Washington sets deployment conditions on the labs that supply the UK market, those conditions become the operative rules for British buyers regardless of what Westminster does, in the same way US export controls already decided which UK firms could access Anthropic’s most capable security models. Watching which senators Altman chose is a reasonable proxy for where the industry expects the binding text to originate.