Now, even the biggest AI boosters in the Administration concede that AI-fuelled hacks into infrastructure, biological threats, attacks on nuclear command-and-control systems, job losses and global financial instability seem more possible than before AI agents went rogue.
The issue of AI safety is expected to come to a head this weekend as Treasury Secretary Scott Bessent meetshis Chinese counterpart in New York to set the agenda for Trump’s visit in Washington with President Xi Jinping of China, which starts on Wednesday (local time). Bessent is trying to negotiate an agreement between Trump and Xi that would suggest the two countries are working together on AI safety concerns.
But already, officials say, some of the most expansive ideas being discussed among AI safety experts, including an agreement on safety standards and reviews of forthcoming Chinese and American AI models before public release, are unlikely to be part of any initial accord.
One diplomat involved in the effort said he believed “the best we can hope for” is the creation of some kind of US-China working group that might build on a Biden-era agreement with China to bar the use of artificial intelligence in decisions about the employment of nuclear weapons. But that does not begin to get at the complexity of the problem of unpredictable AI-driven agents, which took the industry and the Government by surprise this summer.
Nor does it get to the fundamental issues surrounding imposing safeguards on frontier models. The US could go its own way and impose mandatory reviews of the safety of new, American-generated models, but Trump and his aides believe that would effectively drive the global market to Chinese largely open-source alternatives. And because those open-source models can be modified by users – who could remove the “guardrails” meant to limit their powers – a safety review may mean little.
Trump wants to meet AI executives next week, according to company representatives, although a White House official on Thursday (local time) said nothing was on the schedule.
If that session happens, it would presumably include some of the AI executives who, in recent days, have been calling for the kind of slowdown in the release of new models that Trump says he opposes.
Few expect Dario Amodei, a founder of Anthropic, to be among the invitees: his company’s products have been banned by the Pentagon and he appeared to anger Trump by proposing last weekend a detailed plan for the industry to pace itself and bring in independent safety inspectors. Trump dismissed him, and denounced an unnamed class of “conspiracy theorists, treasonists, traitors and leakers” who were politicising the debate over AI and undercutting American leadership.
But another Anthropic founder, Tom Brown, who raises less ire in the White House, talks to officials regularly. In recent days, Trump has also talked with Sam Altman of OpenAI and Elon Musk of SpaceX, both of whom endorsed Amodei’s plan to slow down, test new models for vulnerabilities and let in outside inspectors, at least in general terms.
Bessent and the White House chief of staff, Susie Wiles, have been increasingly concerned over the past several months that AI could cause massive disruptions across industries, including hacks into major financial institutions that could destabilise the world economy.
Wiles this year began organising an informal group to think about how to deal with the risk created by new, more powerful models. Bessent is chief among the members of that informal group, an Administration official said, but other experts are called in depending on the topic.
The White House pointed to Wiles’ social media post on May 6 as the foundational statement of how Trump balances the risks and opportunities of the new technology. “This administration has one goal: ensure the best and safest tech is deployed rapidly to defeat any and all threats,” she wrote. She said that “the White House will continue to lead an America First effort that empowers America’s great innovators, not bureaucracy, to drive safe deployment of powerful technologies while keeping America safe”.
Asked about Trump’s accusation that AI risk is overblown, White House spokeswoman Elizabeth Huston said: “President Trump is saying that Democrats using AI concern as a way to score points is a hoax, just like their claims that climate change would kill everyone in six years”.
She added that Trump had reversed the “failed approach” of the Biden Administration, which she said had “stifled innovation, hurt consumers and nearly cost America its competitive edge over China”.
But at a moment when AI has emerged as arguably the most important long-term issue for the country’s economic health and national security, no single person in the Trump White House is in charge of co-ordinating AI policy. Marco Rubio, as national security adviser, is rarely discussed as a player on these issues. The White House eliminated the post of deputy national security adviser for cyber and emerging technologies created during the Biden Administration.
And the national cyber director, a position created by Congress, has only indirect authority over artificial intelligence issues. The current director is Sean Cairncross, a former senior official at the Republican National Committee, who was light on cyber and AI experience before taking the post.
And there has been a purging of technical expertise, including at the Cybersecurity and Infrastructure Security Agency (Cisa), part of the Department of Homeland Security. The acting director of Cisa said last week that the agency was hiring about 250 new employees after shedding about 1000 jobs – roughly one-third of its workforce – since Trump returned to the White House.
As a result, AI has become the Administration’s latest hot potato, with various officials dipping in and out of influence depending on the topic, or the day of the week.
Still, other officials, particularly in the much-shrunken National Security Council and the Energy Department, share Amodei’s fears about charging ahead with new, more powerful models amid growing evidence that they are increasingly unpredictable.
Amid the rising turmoil over AI, several White House staff members working on security and policy issues related to the technology have left in recent months.
The cyber director’s office, which has been involved in AI safety debates, had two of its top policy officials depart this summer. Sriram Krishnan, the White House’s senior policy adviser for AI, also left. And just this week, Sihao Huang, a top AI policy official in the White House’s Office of Science and Technology Policy, announced on social platform X (formerly Twitter) that he, too, was leaving – to join Anthropic.
Former Administration officials said that working on AI policy has been a frustrating endeavour because of Trump’s deference to tech evangelists and investors including Sacks, who are deeply sceptical of imposing additional guardrails or regulation.
And they worry about embracing any ideas for regulation, especially after Sacks and others persuaded the President at the last minute to pull back on an executive order this year that would have required mandatory government reviews of new AI models. It was replaced with a voluntary review regime.
Trump seems driven chiefly by the sense that AI is the way to economic growth and that nothing can be allowed to stand in its way.
He has long promised that his presidencies would bring on a manufacturing boom that would usher in a new “golden age” for America. His first-term promise was that he would achieve that by reviving heavy industries, claiming his policies would bring coal mining jobs back to the deindustrialised Midwest.
But those jobs did not materialise. When he returned to the presidency for his second term, Trump believed the tech companies’ gargantuan capital expenditures could be the necessary fuel to run the economy hot.
He latched on to this idea during the late campaign season. In a private meeting with OpenAI executives in June 2024, Trump discussed creating a version of “Operation Warp Speed” – his all-of-Government enterprise to fast-track Covid-19 vaccines – for building data centres.
Ever since, he has been determined to fuel AI growth at nearly any political cost, to keep alive what he has referred to as “the golden goose”. Trump, a real estate developer, is eager to keep building, dismissing the concerns of states and towns that want to slow or halt construction.
“The robots will not be taking over,” Trump insisted when he called into a large conference where Sacks and Jensen Huang were on stage this week. “The AI will not be taking over the rest of the world.”
This article originally appeared in The New York Times.
Written by: David E. Sanger, Jonathan Swan, Cecilia Kang and Dustin Volz
Photographs by: Haiyun Jiang and Manuel Orbegozo
©2026 THE NEW YORK TIMES
