US Revokes 90-Day Federal Review Order for Frontier AI Models, Highlighting Regulatory Divergence Among US, China, and EU

Policy Shift: Inside the Review Order Withdrawal

On May 24, 2026, the US administrative measure that would have required a 90-day federal review before releasing frontier AI models was withdrawn hours before signing. Official documents reveal the core reason was fear that such a review would slow America's technological lead in the AI race against China. Following the complete repeal of the Biden-era Executive Order 14110 by Trump, the US still lacks a unified federal AI regulatory framework at the federal level.

Political Dynamics: Winners and Losers

Labs emphasizing open research, such as Meta and Google DeepMind, became direct beneficiaries, as they can continue rapid model iteration without waiting for federal approval. OpenAI and Anthropic face reduced short-term compliance costs, but their safety-first stance is undermined. EU regulators and China's Cyberspace Administration are indirect beneficiaries, as the US regulatory vacuum makes global companies more inclined to accept their existing rules. Safety advocacy groups and some Democratic members of Congress are clearly disadvantaged, as the federal review mechanism they pushed for has completely fallen through.

Comparison of Regulatory Paths: US, China, and EU

The US adopts a post-hoc, market-driven model, relying on existing antitrust and export control tools, prioritizing technological leadership. The EU's AI Act, now in force, mandates pre-deployment impact assessments and transparency reports for high-risk systems, creating strict compliance barriers. Since 2024, China has implemented a large model filing system requiring pre-launch safety assessments, focusing on content control and national security. This US withdrawal further widens the gap among the three, forcing multinational labs to simultaneously navigate three distinct compliance paths.

Short-Term Impact on AI Labs

In the short term, OpenAI, Anthropic, and others can immediately proceed with internal testing and limited releases of GPT-5-level models or subsequent versions of Claude, without an extra 90-day wait. Meta's Llama series open-source plans will also accelerate, reducing risks of funding and talent loss due to regulatory uncertainty. However, labs must still establish their own internal safety assessment processes to address potential state-level legislation and international litigation.

Medium-Term Impact on AI Labs

In the medium term, around 2027, the EU and China will require US labs to submit detailed reports and filings for high-risk systems, potentially raising compliance costs by over 30%. The lack of a unified federal framework subjects US labs to dual pressure from fragmented state regulations and export controls, forcing them to set up compliance subsidiaries overseas or adjust model architectures to meet multiple requirements. Overall, this withdrawal reinforces the US's corporate self-governance regulatory style but also increases the long-term risk of global AI governance fragmentation.