AI Regulation: Brussels blinks, DOJ sues a state, Connecticut moves  May 13th 2026

Watch & Listen First

• Scaling Laws — An EU Perspective on America's Approach to AI, with Marietje Schaake — Stanford's Schaake on how Brussels reads the Trump AI framework. Released May 1.
• Lawfare Daily — Former FCC Chair Tom Wheeler on AI Regulation — Wheeler argues a new agency is required; direct response to the White House framework.
• European Parliament press debrief — AI Act negotiations — Co-rapporteurs explain the omnibus trade-offs on the record.

Key Takeaways

• High-risk EU obligations slip to December 2027. Annex III standalone systems get a 16-month delay; Annex I embedded products move to August 2028. The watermarking grace period was shortened to three months, deadline December 2, 2026.
• Federal preemption is now litigation, not rhetoric. DOJ's intervention in x.AI v. Weiser is the first time Washington has sued a state to kill an AI law. Expect the same playbook against California, Connecticut, and New York.
• Connecticut SB 5 sets the new state ceiling. Companion-bot disclosure, provenance watermarking for >1M MAU generative providers, ADM rules for employers. First provisions effective October 1, 2026.
• Bartz v. Anthropic fairness hearing on May 14 is the AI copyright milestone of the year. $1.5B settlement, 91.3% claim rate. The order shape will be the template plaintiff's firms copy next.
• The omnibus is redistribution, not deregulation. Brussels added a hard EU-wide ban on non-consensual intimate-image generation while delaying high-risk timelines.

The Big Story

EU Council and Parliament agree to delay high-risk AI Act rules to 2027 in omnibus deal · May 7, 2026 · European Commission
-> The provisional Digital Omnibus shifts Annex III standalone high-risk obligations to December 2, 2027 and Annex I embedded products to August 2, 2028 — pinned to Commission certification that harmonised standards exist. Centralized enforcement at the AI Office expands, SME relief extends to small mid-caps, sandbox deadline slides to August 2, 2027. Safety win: a hard ban on AI generating non-consensual sexually explicit content, providers and deployers, December 2, 2026 compliance (TechPolicy.Press). The GPAI Code of Practice survives, but every downstream milestone moved.

Also This Week

DOJ intervenes in xAI suit challenging Colorado AI Act; court suspends enforcement · April 24-27, 2026 · Department of Justice
-> First-ever federal intervention against a state AI law — DOJ's AI Litigation Task Force, stood up under the December 2025 executive order, is now the lead enforcement vehicle, and the joint stay of Colorado's June 30 effective date previews California and Connecticut suits to come.

Connecticut General Assembly passes SB 5, governor expected to sign · May 1, 2026 · CT Mirror
-> The Artificial Intelligence Responsibility and Transparency Act clears the House 131-17 and Senate 32-4; covers companion-bot disclosure, ADM bias audits for employment, and provenance metadata for generative providers above one million MAU — first provisions effective October 1.

White House studying AI security executive order, FDA-style pre-release review on the table · May 8, 2026 · Federal News Network
-> A senior White House official tells reporters the administration is drafting a process to vet frontier models for cyber risk before public release — a reversal from the light-touch March framework and a response to the Mythos breach pattern.

Five publishers and Scott Turow sue Meta over LLaMA training data · May 5, 2026 · Association of American Publishers
-> Hachette, Macmillan, McGraw Hill, Elsevier, and Cengage allege Meta and Zuckerberg personally directed use of pirated books for model training; complaint cites Bartz v. Anthropic as the damages floor.

Deadlines & Compliance

• May 14, 2026: Bartz v. Anthropic final approval hearing, N.D. Cal. Judge rules on the $1.5B settlement structure (Publishers Weekly).
• May 31, 2026: GSA AI procurement clause (GSAR 552.239-7001) comment period closes (Lawfare).
• June 30, 2026: Colorado AI Act original effective date — stayed pending x.AI v. Weiser and possible repeal via SB 26-189.
• October 1, 2026: Connecticut SB 5 first-tranche provisions (companion-bot disclosure, generative-AI transparency) assuming Lamont signs (Transparency Coalition).
• December 2, 2026: EU AI Act watermarking compliance deadline and non-consensual intimate-content prohibition compliance date.

Worth Reading

• White House AI Framework Proposes Industry-Friendly Legislation — Lawfare's structural read on the March framework that became this month's litigation strategy; required for anyone drafting state-law compliance memos.
• Mythos Fallout: U.S. Government Weighs AI Model Regulation — Connects the cybersecurity-driven pivot toward pre-release model review with the broader White House posture shift reported May 8.

Brussels delayed, Washington sued, Hartford legislated. Compliance teams that win the next quarter rebuild their calendars around new dates, not old ones.