Regulators move hard on AI
The U.S. Senate released a draft national AI framework this week that would pre-empt state laws and centralize oversight — a potential stress point for firms used to patchwork rules. At the same time Colorado’s task force pushed legislative ideas that would leave liability questions to courts and create a three‑year safe harbor for supervised AI legal tools, and the EU is advancing an outright ban on AI systems that generate non‑consensual sexual images after high‑profile abuses. (news.bgov.com) (denverpost.com) (abajournal.com) (bloomberg.com)
Sen. Marsha Blackburn rolled out a discussion draft called the TRUMP AMERICA AI Act on March 18, 2026, packaging previously introduced measures including the Kids Online Safety Act and the NO FAKES Act into one federal framework. (blackburn.senate.gov)) The draft tasks the Federal Trade Commission with promulgating minimum “reasonable safeguards” and places an explicit duty of care on AI developers to assess and mitigate foreseeable harms. (reason.com)) It obliges large “frontier” model developers to adopt catastrophic‑risk protocols, publish transparency reports and report incidents to the Department of Homeland Security, while carving out preemption authorities for certain federal AI rules. (reason.com)) The proposal also seeks to alter Section 230 liability protections, mandate third‑party audits and formalize regular risk‑assessment and model‑transparency obligations for covered platforms. (blackburn.senate.gov)) Legal analysts warn the federal preemption strategy faces constitutional and political hurdles, and Senate backers acknowledge the bill will need bipartisan support amid a compressed 2026 calendar. (ropesgray.com)) In Colorado, the governor‑backed AI Policy Work Group agreed to a draft to repeal and replace parts of the 2024 law and push the implementation timetable (previously moved) toward a later effective date, as lawmakers negotiate scope and enforcement. (coloradopolitics.com)) Separately, Colorado’s attorney regulation counsel adopted a three‑year nonprosecution policy for developers of legal‑help AI that requires lawyer supervision and clear consumer disclosures, and EU institutions this month advanced amendments to ban AI systems that generate non‑consensual sexualized or child sexual‑abuse images while delaying some high‑risk AI deadlines. (abajournal.com))
