The Trump administration is considering a new round of guardrails for advanced AI models, according to The Information. The report signals a notable shift for a White House that spent its early months tearing down the Biden-era AI rulebook and pitching itself as the deregulation administration on frontier tech.
This is significant because it suggests Washington is rethinking its hands-off posture as model capabilities accelerate. After scrapping Biden’s October 2023 executive order on safe and trustworthy AI, the administration leaned heavily on its AI Action Plan, which prioritized speed, compute buildout, and American competitiveness over Europe-style rules. Now, per The Information, officials are weighing what specific limits should apply to the most powerful models.
What’s reportedly on the table
The Information’s reporting points to renewed federal interest in setting boundaries for frontier systems. While the full scope isn’t public, policy debates in Washington this year have circled around a familiar set of levers:
- Pre-deployment testing for models above certain capability thresholds
- Reporting requirements for training runs that cross compute or risk benchmarks
- Export controls on model weights and the chips that train them
- Safety disclosures tied to dual-use capabilities like cyber and bio risks
Any new guardrails would likely route through the Commerce Department and the AI Safety Institute, the body Biden stood up at NIST and that the Trump team has so far kept alive under a slightly different banner.
How we got here
The status quo before this shift was clear. The administration killed Biden’s executive order on day one. It pushed federal agencies to clear regulatory underbrush. It signed off on massive infrastructure deals tied to OpenAI, Oracle, and others. The message to labs was simple: build faster, ship faster, beat China.
What changed? A few pressures stacked up at once. Frontier labs keep releasing models that score higher on dangerous-capability evals. State legislatures, especially California and New York, are filling the federal vacuum with their own AI bills, creating the patchwork industry says it wants to avoid. And national security hawks inside the administration have grown louder about model weights leaking to adversaries.
A federal floor starts to look more attractive when the alternative is fifty different state regimes plus an unbounded export risk.
Why it matters for the industry
For frontier labs, this is the policy question of the year. New federal guardrails could lock in compliance costs that favor incumbents like OpenAI, Anthropic, and Google, since they already run internal safety evaluations and have policy teams ready to engage. Smaller labs and open-weight model providers would feel the squeeze harder.
For enterprises building on top of these models, any reporting or testing regime adds friction at the API layer. Expect provider terms of service to tighten if Washington sets new disclosure rules. Procurement teams should start asking vendors how they’d handle a federal pre-deployment testing requirement, because the answer shapes deployment timelines.
For the open-source community, the stakes are highest. The fight over whether to treat model weights as controlled exports has been simmering since the original Biden order, and a Trump-branded version of those controls would be much harder for the open-weight camp to push back on politically.
What to watch next
Watch for three signals in the coming weeks. First, any executive action or NIST framework update that codifies what counts as a frontier model. Second, statements from David Sacks, the White House AI and crypto czar, who has been the loudest internal voice on letting the market run. Third, how labs respond, since their public comments often telegraph what’s already been negotiated behind closed doors.
The Trump administration’s AI strategy has been defined so far by what it removed. If The Information’s reporting holds, the next chapter is about what it adds back.
Full details are available at the original source.