Trump Administration Shifts Course on AI Regulation
The Trump administration is considering tighter government oversight of artificial intelligence models, marking a reversal from its initial hands-off stance. The White House is weighing an executive order that would require formal government review of AI models before their public release, according to reporting from the New York Times citing U.S. officials.
The shift comes less than four months after President Trump revoked a 2023 Biden-era executive order on AI regulation, calling it an obstacle to innovation. At the time, the administration signaled it would let tech companies develop AI with minimal government interference.
Voluntary agreements take shape
Rather than wait for a formal executive order, the Department of Commerce announced Tuesday that it has reached voluntary agreements with Google DeepMind, Microsoft, and xAI. Under these deals, the Center for AI Standards and Innovation will conduct "pre-deployment evaluations" of the companies' models before they reach the public.
A White House official told MarketWatch that discussion of a potential executive order is "speculation" and that any policy announcement will come from the president.
What triggered the reversal
The pivot appears tied to national security concerns. Anthropic released a model called Mythos last month that can discover and exploit cybersecurity vulnerabilities. The company withheld public release but shared it with select organizations through Project Glasswing.
James Lewis, a senior adviser at the Center for Strategic and International Studies, said the administration's alarm over the model's capabilities likely drove the policy shift. "I think a lot of it comes from Anthropic persuading the Defense Department that this was a risk that needed to be seriously considered," Lewis said.
David Sacks, Trump's top AI adviser who had championed deregulation, left the White House in March. His departure may have cleared the path for the administration to move toward more oversight.
How far will regulation go?
Lewis expects the Trump administration will focus narrowly on national security rather than adopt the Biden approach, which addressed safety, privacy, equity, and diversity in AI. "This administration is a little allergic to government surveillance, so I doubt an executive order would go beyond national security," he said.
The CAISI partnership remains voluntary, meaning companies cannot be forced to participate. Lewis suggested the program could offer legal protections similar to the Cybersecurity Information Sharing Act, giving companies that submit models for review safe harbor from prosecution.
Industry concerns
Tech advocates worry that pre-approval requirements could slow U.S. AI development. Daniel Castro, president of the Information Technology and Innovation Foundation, said a sweeping review regime would force companies to "seek permission to innovate, slowing development to the pace of Washington rather than the market."
Castro also noted that countries without similar restrictions could advance faster, and open-source models released abroad would face no constraints. This could undercut the U.S. AI ecosystem's competitive position.
For managers overseeing AI initiatives, the policy shift means increased scrutiny of model development timelines and deployment processes. Understanding both the voluntary CAISI program and potential future requirements will be essential for planning.
AI for Executives & Strategy resources can help leaders navigate these emerging regulatory expectations.
Your membership also unlocks: