White House briefs AI firms on executive order for model review
The White House is moving ahead with a planned executive order that would establish a voluntary framework for government review of advanced artificial intelligence models before their public release, according to people familiar with the matter. The Office of the National Cyber Director hosted a briefing on Tuesday with leading AI companies including OpenAI, Anthropic and Reflection AI, the Information reported Wednesday.
President Donald Trump could sign the executive order as soon as Thursday, the report said. The order would empower intelligence agencies and other government bodies to review frontier AI models up to 90 days before companies make them available to the public. Companies participating in the voluntary program would share advanced models with agencies and also grant pre-release access to critical infrastructure providers such as banks, one person familiar with the planning told Reuters.
The White House was working to get AI company CEOs to a signing ceremony with President Trump, another source familiar with the planning said.
What the order would do
Under the proposed framework, developers of frontier AI models would notify the U.S. government ahead of major releases. The companies could share their models with government agencies as early as 90 days before public launch, giving officials time to assess potential risks. The framework is described as voluntary, meaning companies would not face legal penalties for opting out.
OpenAI and Reflection AI did not immediately respond to requests for comment from Reuters. Anthropic declined to comment.
Why the order is coming now: Security fears among Trump supporters
Pressure for stricter AI oversight has been building within parts of Trump's political base. MAGA activists, including former Trump adviser Steve Bannon and right-wing political organizer Amy Kremer, have been pressing the White House to require AI developers to submit their most capable models for government security tests. Their concerns grew louder after the release of Anthropic's Mythos model, which raised new questions about safety and misuse potential.
On the other side of the debate are tech industry supporters such as venture capitalist Marc Andreessen and former Trump AI official David Sacks, who have resisted mandatory regulations. Sacks stepped down as Trump's lead AI official in March and now co-chairs the president's tech advisory committee. Trump's AI policies in his second term have largely reflected the industry's preference for light-touch approaches.
The executive order represents a middle ground, say people who have seen drafts. It stops short of forcing companies to submit models, but the 90-day review window gives the government time to issue warnings if it finds vulnerabilities.
National security dimensions
National Cyber Director Sean Cairncross, who serves as Trump's principal adviser on cybersecurity policy and strategy, played a key role in developing the order. The National Security Agency, which often works with intelligence agencies on AI threats, directed questions to the White House when asked about specific provisions.
A White House spokesperson called any discussion about AI policy details "speculation" ahead of the expected signing.
The debate over voluntary versus mandatory rules
The clash between MAGA activists and tech industry backers reflects a broader tension within the Trump coalition. Bannon and Kremer have argued that voluntary frameworks are insufficient to prevent AI-enabled cyberattacks, disinformation campaigns or other national security threats. They want mandatory testing requirements with clear enforcement mechanisms.
Tech industry representatives counter that heavy-handed regulation would stifle innovation and drive AI development overseas, particularly to China. The White House has tried to balance these competing interests, with Chief of Staff Susie Wiles, Science and Technology Adviser Michael Kratsios, Wiles' deputy Walker Barrett, and Cairncross working with AI companies over the past month.
What industry leaders are saying
OpenAI has publicly supported voluntary cooperation with government while opposing binding rules. Anthropic has taken a more cautious stance, advocating for pre-release testing but not necessarily for government mandates. Reflection AI, a newer player in the frontier model space, has not taken a public position on the order.
A spokesperson for the tech advisory committee, which includes representatives from major AI companies, said the group supports the voluntary approach as a starting point but expects further discussions as models become more capable.
What the 90-day review window could mean for model rollouts
If companies choose to participate, the review period could change how and when new AI models reach the market. A 90-day delay between completing a model and releasing it to the public would give intelligence agencies time to test for vulnerabilities, potential misuse, or alignment with cybersecurity standards.
For companies working on cutting-edge models, that window could affect competitive positioning. A delay might allow rivals to catch up or give governments time to prepare countermeasures if a model poses risks.
Implications for critical infrastructure
The order asks developers to give pre-publication access to critical infrastructure providers such as banks, energy grids and transportation systems. This would allow those entities to update their defenses against new AI-driven threats before the technology becomes widely available.
However, the voluntary nature of the framework means infrastructure providers might not see the most dangerous models if companies choose not to participate. Critics argue that critical infrastructure protection requires mandatory reporting.
Broader implications for AI policy and global competition
The executive order comes at a time of heightened global AI rivalry. China's Manus founders have reportedly sought $1 billion to undo a Meta takeover, and Nvidia's Jensen Huang has conceded that Huawei won China's AI chip market. These developments underscore the stakes for U.S. leadership in AI.Learn more about Huawei's dominance in China's AI chip market
The voluntary framework aligns with the approach taken by many other countries, including the United Kingdom's AI Safety Institute, which relies on company cooperation rather than legal mandates. However, the European Union's AI Act takes a more stringent approach, imposing binding obligations on high-risk models.
What this changes
If signed on Thursday, the order would formalize a process that until now has been carried out on an ad hoc basis. Some AI companies already provide early access to government researchers, but the new framework would standardize the review window and notification requirements.
The order also signals that the Trump administration, despite its pro-industry leanings, is listening to security concerns from its base. Whether the voluntary approach satisfies MAGA activists or leads to demands for stronger action remains an open question.
The tech industry, for its part, may see the order as a manageable compromise — at least for now. As models continue to advance, the balance between innovation and oversight is likely to shift.
Looking ahead: The next frontier of AI governance
The executive order is unlikely to be the final word on AI oversight in the United States. Trump's AI policies have evolved significantly since his first term, when the focus was on removing regulatory barriers. The emergence of highly capable models like Mythos has shifted the conversation toward security, even among some of the industry's strongest supporters.
Congress has shown little appetite for comprehensive AI legislation, leaving the White House to act through executive orders. The voluntary framework could serve as a template for future regulation, or it could become a point of contention if a major AI incident occurs.
The human element
The debate over AI oversight takes place against a backdrop of job displacement and public skepticism. A recent survey found that young people are "booing, not applauding" the arrival of AI bots in the workplace. Standard Chartered announced plans to cut over 7,000 jobs and boost AI to replace "lower-value human capital." Major financial institutions are rolling out AI tools, while urging employees to embrace change.
These trends highlight why AI governance matters beyond Washington. The decisions made through executive orders and industry frameworks will shape not only national security but also employment, privacy and the distribution of power in the digital economy.
The White House executive order, if it proceeds as expected, represents a step toward institutionalizing oversight of the most powerful AI systems. Whether it goes far enough — or too far — will depend on the next model release, and on how the voluntary framework holds up under pressure.
Comments