US AI Policy Centralising
Recent legal and policy analysis says the emerging U.S. AI framework is centralising authority and emphasising federal pre‑emption, even though detailed, binding rules are still missing. (mondaq.com) A separate assessment from the Bloomsbury Intelligence and Security Institute describes the framework as favouring deregulation and federal supremacy while leaving companies uncertain about operational obligations. (bisi.org.uk)
The United States is moving toward one national artificial intelligence policy, even as Congress still has not passed a detailed federal rulebook. (federalregister.gov) President Donald Trump signed Executive Order 14365 on December 11, 2025, directing his administration to build a “minimally burdensome” national framework and seek federal rules that preempt conflicting state artificial intelligence laws. The White House followed on March 20, 2026, with a legislative framework for Congress. (federalregister.gov) (whitehouse.gov) The administration has also shifted its main federal artificial intelligence office toward standards and industry collaboration. On June 3, 2025, the Commerce Department renamed the United States Artificial Intelligence Safety Institute as the Center for AI Standards and Innovation, and the National Institute of Standards and Technology says the center will help develop voluntary standards and testing. (content.govdelivery.com) (nist.gov) That federal push arrived as states kept writing their own laws. The National Conference of State Legislatures says all 50 states introduced artificial intelligence bills in 2025, and 38 states adopted or enacted about 100 measures. (ncsl.org) Colorado’s law shows what Washington is trying to head off. Colorado Senate Bill 24-205 requires developers and deployers of “high-risk” artificial intelligence systems to use reasonable care against algorithmic discrimination, and its effective date was moved to June 30, 2026. (leg.colorado.gov) (paulhastings.com) California has taken a different route, passing narrower laws on specific harms. Governor Gavin Newsom announced on September 19, 2024, that he signed bills targeting sexually explicit deepfakes and requiring some artificial intelligence watermarking, and California’s Artificial Intelligence Transparency Act was chaptered the same day. (gov.ca.gov) (legiscan.com) Congress has already tested preemption and pulled back. A House bill introduced on September 16, 2025, proposed a temporary moratorium on some state artificial intelligence laws, but later analyses say a broader 2025 Senate effort to block state enforcement failed. (congress.gov) (labs.cloudsecurityalliance.org) State officials have pushed back directly. A coalition of attorneys general wrote congressional leaders on May 15, 2025, opposing a proposed 10-year ban on state enforcement of artificial intelligence laws and warning it would override state consumer-protection rules. (scag.gov) Legal and policy analysts now describe the result as centralized in direction but thin on binding obligations. Mondaq said the executive order showed why artificial intelligence governance “requires centralisation,” while the Bloomsbury Intelligence and Security Institute said the framework favors federal supremacy and deregulation while leaving firms unsure what, exactly, they must do. (mondaq.com) (bisi.org.uk) For companies, that means the map is still split. Washington is asserting control from the top, but until Congress turns that framework into statute, businesses still have to track the state laws already on the books. (whitehouse.gov) (ncsl.org)
