The White House is considering a policy that would require artificial intelligence models to be vetted by the federal government before they are released to the public, according to a CNBC report by Kate Rooney. The report did not specify which models would be covered, what the vetting process would look like, or whether the plan would require new legislation or rely on existing executive authority. No timeline was provided, and the proposal remains in an early stage of deliberation.

If enacted, the measure would mark a significant escalation in the U.S. government's approach to AI oversight. Until now, federal action on AI has largely focused on voluntary commitments, guidelines, and after-the-fact enforcement. The Biden administration's October 2023 executive order on AI, for example, required developers of the most powerful models to share safety test results with the government after training but before release. A pre-release vetting requirement would go further, giving regulators a direct gatekeeping role over which AI products can enter the market.

The concept is not entirely new. The European Union's AI Act, approved earlier this year, imposes different levels of scrutiny depending on risk, including mandatory conformity assessments for high-risk systems. China already requires approval for certain generative AI products. The U.S. has so far resisted such a framework, arguing that heavy-handed regulation could stifle innovation and cede advantage to global competitors. A pre-release review requirement would represent a shift in that posture.

The CNBC report did not name which White House office or agency might oversee the vetting process. It is unclear whether the National Institute of Standards and Technology, the Federal Trade Commission, or a new entity would be tasked with evaluating models. The report also did not clarify what criteria would be used to determine which models are subject to review. Thresholds based on computing power, training data size, or intended use case are possible, but none were confirmed.

Proponents of pre-release review argue that AI models, particularly large language models and multimodal systems, can generate harmful content, propagate bias, be used for disinformation, or enable cyberattacks. Catching these problems before deployment, they say, is more effective than trying to recall or patch them after release. Critics counter that vetting could slow down innovation, create bottlenecks at review agencies, and give incumbent companies an advantage over startups that cannot afford long approval cycles. They also question whether government reviewers would have the technical expertise to evaluate models, especially as capabilities evolve rapidly.

The CNBC report did not detail how the White House is weighing these trade-offs. It also did not mention whether the proposal has support or opposition within the administration. The report's primary source appears to be unnamed officials familiar with the discussions.

If the plan moves forward, it would likely face legal and practical challenges. The First Amendment could be raised as a barrier, since some AI-generated output might be considered protected speech. Constitutional questions about prior restraint would almost certainly be litigated. Congress would also need to weigh in, either through appropriations—funding the review agency—or via new statutory authority. The report did not indicate whether the administration intends to pursue legislation or use executive orders alone.

The news comes as the global conversation around AI safety intensifies. In March 2024, the United Nations adopted a resolution on safe AI, and several countries have established AI safety institutes. The U.S. AI Safety Institute, housed within NIST, was created in late 2023 but has focused on testing existing models rather than approving new ones. Whether that institute would be expanded to perform pre-release reviews is unclear from the CNBC report.

For now, the White House's consideration of such a policy signals that the government is moving from voluntary to mandatory approaches. Developers building the next generation of AI systems should expect that the regulatory environment is not static. The absence of detailed proposals in the CNBC report suggests that the administration is still building its case internally. Public consultations, draft frameworks, or pilot programs could emerge before any formal rule is proposed.

What remains unknown is whether the vetting requirement would apply to open-source models, which can be downloaded and modified by anyone. Several recent advances in AI have come from open-source projects, and regulating them poses unique challenges. The CNBC report did not address this distinction.

Ultimately, the story is less about a concrete plan and more about the direction of travel. The White House is actively exploring a mechanism that would give the government a say in which AI models can be released. That alone is a notable departure from the hands-off approach that has characterized U.S. policy toward the technology so far.

As the details remain sparse, developers, researchers, and the broader tech industry will be watching closely for any announcement of a formal proposal. The CNBC report is the first public indication that such a change is under active consideration at the highest level of the U.S. government.