Skip to content
Policy

White House Keeps Its Frontier AI Model Review Framework Private and Voluntary

Cyber director Sean Cairncross says AI labs already hand over new models for government evaluation and that a strict regulatory regime would be too slow.

·1 min read

Prefer Nakoda News on Google

White House cyber director Sean Cairncross spoke at Semafor's Silicon Valley & The World event on Monday.
White House cyber director Sean Cairncross spoke at Semafor's Silicon Valley & The World event on Monday.Photo: Semafor

The Trump administration has no plans to publish its voluntary framework for reviewing new advanced AI models, or to require anyone to comply with it, White House cyber director Sean Cairncross said on Monday, Semafor reported.

Speaking at Semafor's Silicon Valley & The World event, Cairncross said frontier labs have cooperated with the government. "The labs have been extremely good about being in communication with us and giving their new models for evaluation to be assessed," he said. Putting a strict regulatory regime in place "isn't going to work at the speed that we need," he added.

An executive order set the framework in motion in June, but its full text has not been released. Cairncross said the administration held it back to allow a "frank conversation" with AI companies. He said he believed the White House would soon make "it much more clear how this is working in practice."

The White House has taken a lighter-touch approach to regulating the AI industry than Europe or the Biden administration. It has also taken steps to block states from stepping in with their own rules.

Cairncross was upbeat about AI's value for cyber defence, describing its capabilities as "unimaginable even a short time ago."

Estonian Prime Minister Kristen Michal, who appeared on stage alongside him, said Russia's hacking capabilities have improved along with AI over the past year. He said Estonia's highly wired government has come under increasing attack, and blamed private actors as well as Russia, North Korea and Iran.

Asked whether cyberattacks could be considered acts of war, Michal said he believes some could rise to NATO's definition of an "attack" under Article 5 of the alliance treaty. "It's mainly a question about the consequences," he said.

By

Nakoda News Policy Desk

Regulation, government strategy and the rules taking shape for AI and technology.

About the desk

Related coverage