Key Takeaway
According to the New York Times citing US officials, the Trump administration is discussing establishing an AI model pre-release review mechanism through executive order — requiring powerful AI models to undergo government review before public release. This represents a 180-degree reversal from the Trump administration previous stance of “mass revocation of AI regulatory constraints to free AI innovation” and could fundamentally change the release pace and competitive landscape of AI models in the US and globally.
Event Timeline
| Time | Event |
|---|---|
| January 2025 | Trump signs executive order, revoking Biden-era AI regulatory constraints |
| April-May 2026 | White House internally discusses new AI model review mechanism |
| May 4, 2026 | NYT first reports, sources say executive order “may be in the pipeline” |
In just 16 months, the stance shifted from “comprehensive deregulation” to “pre-release review” — a change in speed and magnitude that exceeds market expectations.
Core Contents of the Review Mechanism
Based on disclosed information, potential plans include:
- Formal government review process: AI working group members composed of tech industry executives and government officials, reviewing potential risks of new models
- Pre-release approval: Powerful models must obtain government approval before public release
- De facto licensing system: Critics point out this essentially establishes a “license” system for AI models
Reactions from Various Parties
Industry concerns: Commentators note that any “pre-release review” could be tantamount to a de facto licensing system, and implementing it through executive order rather than legislation is even more concerning. This could lead to:
- Model release delays, affecting US AI competitiveness
- Small startups being excluded due to compliance costs
- Impact on the open-source model community
Logic behind the policy reversal: Several driving factors may be behind the shift:
- Cases of Chinese models like DeepSeek bypassing safety restrictions through “distillation” raising concerns
- Rapid evolution of AI Agent capabilities, increasing potential risks of autonomous task execution
- Election cycle policy pressures
Impact on US-China AI Competition
The impact of this policy reversal on US-China AI competition is two-way:
Short-term negative for US: The review mechanism will slow down US model release pace. With Chinese models (DeepSeek V4, Qwen 3.6, Kimi K2.6) already globally competitive, the time window is extremely valuable.
Long-term complexity: If the review mechanism improves the credibility and security of US AI systems, it could gain advantages in the international market. But only if the review process is efficient rather than bureaucratic.
Impact on open source: If the review covers open-source model releases, it will directly impact the iteration speed of ecosystems like Llama and Qwen. This is the greatest concern of the global open-source community.
Relationship with Existing Regulatory Frameworks
| Regulatory Framework | Scope | Relationship with This Review |
|---|---|---|
| EU AI Act | Deployed within EU | Independent but may converge |
| Colorado Algorithmic Discrimination Act (effective June 30, 2026) | State level | Complementary |
| Previous White House executive order (deregulation) | Federal | Replaced by new policy |
| US State Dept diplomatic cable on Chinese AI distillation | Diplomatic level | Same security concern |
Action Recommendations
- AI Companies: Prepare model safety assessment documentation in advance, establish internal compliance processes even before the executive order is formally issued
- Open-source projects: Monitor whether the executive order draft covers open-source models; may need to adjust release strategies
- Investors: The review mechanism may benefit compliance-focused AI companies while hurting startups relying on rapid iteration
- Developers: If open-source model releases are restricted, the importance of local deployment and fine-tuning will further increase