The Trump administration is considering government oversight of new artificial intelligence models before their public release, marking a potential shift from its earlier hands-off approach.

White House officials have discussed an executive order that would establish an AI working group comprising tech executives and government representatives to develop oversight procedures. Among the ideas under consideration is a formal review process, similar to one in Britain, where government agencies assess models for safety standards without blocking their release. The group would determine which agencies, such as the National Security Agency or the Office of the National Cyber Director, would conduct the evaluations.

These discussions occurred amid meetings last week with executives from Anthropic, Google, and OpenAI. Anthropic CEO Dario Amodei met with White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent last month. The talks followed the April announcement of Anthropic's advanced model, Mythos, which excels at detecting software vulnerabilities but has not been released publicly due to its potency.

The consideration stems from national security concerns, particularly the risk of AI-enabled cyberattacks. Pentagon and intelligence officials are evaluating models like Mythos for cyber defense and military applications. A dispute arose earlier this year when Anthropic rejected Pentagon demands for unrestricted access in a $200 million contract, leading the Defense Department to halt use of its technology in March and designate the company a national security risk. Anthropic filed a lawsuit in response.

Previously, the administration prioritized deregulation to spur innovation and compete with China. President Trump rolled back Biden-era requirements for safety evaluations on models with military potential. Former AI czar David Sacks, who left in March, led those efforts. Trump has described AI as a 'beautiful baby' that must thrive without 'foolish rules,' while Vice President JD Vance emphasized building over safety hand-wringing.

Public apprehension has grown, with a Pew Research Center poll last year finding half of Republicans and slightly more Democrats more concerned than excited about AI's daily impacts. Dean Ball, a former Trump AI adviser, noted the challenge of balancing rapid technology advancement with oversight without overregulating.

Tech leaders have voiced worries that excessive scrutiny could hinder U.S. competitiveness. A White House official described talk of an executive order as speculation, adding that Trump would announce any policy personally. The proposal reflects input from Wiles and Bessent, who have assumed greater roles in AI matters post-Sacks.