Lobbying by Musk, Zuckerberg and Sacks froze a signing ceremony — while TAKE IT DOWN started arresting people.
A week that began with platforms scrambling to meet the federal TAKE IT DOWN Act's 48-hour takedown deadline ended with the White House yanking a long-anticipated AI security executive order off the Resolute desk. In between, the FTC banked its first major settlement of the year over "AI-powered" surveillance ad claims, UK MPs moved to drag generative AI inside the Online Safety Act, and the first federal indictments dropped under the new intimate-deepfake statute. The dominant trend: enforcement is happening on the periphery while Washington stalls at the center.
Watch & Listen First
- Inside the Second International AI Safety Report — CSIS AI Policy Podcast (May 26) — co-authors Stephen Clare and Stephen Casper walk through the report's findings on technical safeguards and the policy implications regulators are already citing.
- WATCH: Trump explains why he postponed signing the AI executive order — PBS NewsHour (May 21) — the President in his own words on why he pulled the order at the last minute.
Key Takeaways
- Federal AI oversight is on hold — again. Trump's pulled EO removes near-term US pressure for pre-deployment model testing; voluntary commitments remain the only game in town.
- TAKE IT DOWN is now operational. Platforms are exposed to $53,088-per-violation FTC penalties, and DOJ has demonstrated it will charge individuals — compliance workflows must be live, not planned.
- "AI-washing" is the FTC's preferred 2026 hook. Cox Media's $930K plus a 20-year monitorship signals deceptive-claim cases will keep coming under existing Section 5 authority.
- The EU clock has not stopped. Despite the Omnibus extension, December 2, 2026 brings both the shortened transparency-labeling deadline and the new nudifier prohibition — earlier than most providers planned for.
- State preemption fight intensifies. With the federal EO shelved, the DOJ AI Litigation Task Force is now the sole active vehicle for the administration's preemption agenda.
The Big Story
Trump delays signing landmark AI security executive order hours before ceremony · May 21, 2026 · Washington Post
-> The order would have directed federal agencies to harden cybersecurity defenses against AI threats and to negotiate voluntary pre-deployment testing of frontier models — a meaningful walk-back of the administration's deregulatory posture, reportedly driven by Anthropic's Mythos cyber-capability disclosures. After overnight calls from Zuckerberg, Musk and AI czar David Sacks, Trump told reporters language "could have been a blocker" to US leadership (TechCrunch, Axios). For builders: the absence of a federal testing regime keeps the December 2025 preemption EO as the operative federal instrument, and leaves state laws — Colorado, TRAIGA, California's ADS — as the binding compliance surface.
Also This Week
Two men charged in first prosecutions under TAKE IT DOWN Act · May 22, 2026 · TIME
-> DOJ's choice to file the inaugural cases against individual creators rather than platforms confirms the criminal-side enforcement model — but platform-side civil exposure under §3 takedown duties begins running concurrently.
FTC settles "Active Listening" AI ad cases against Cox Media Group, MindSift, 1010 Digital Works · May 21, 2026 · FKKS analysis
-> Three orders totaling $930K plus 20-year compliance monitorships establish that vendor-side AI capability claims and consent flows are now first-order Section 5 issues, even when the underlying technology may not have actually worked as advertised.
UK MPs press to extend Online Safety Act to generative AI · May 22, 2026 · Resultsense
-> A cross-party push to bring chatbots and standalone AI companion apps inside Ofcom's remit signals that the UK's "sector-led" posture is fraying, with a government response due this summer.
Northeastern: ChatGPT faces a lawsuit onslaught — can a chatbot be held liable? · May 22, 2026 · Northeastern Global News
-> Two new product-liability suits filed within days — one tied to the 2025 FSU shooting, one to a Texas overdose — extend the Garcia v. Character Technologies "AI as product" theory squarely to OpenAI; expect more plaintiff-side activity until the Northern District of California consolidates.
EU plans to ban non-consensual "nudifier" apps under AI Act amendments · May 19, 2026 · Euronews
-> Implementing detail on the December 2 prohibition is sharpening: providers of any general-purpose model whose outputs are foreseeably misusable will need documented misuse-mitigation by go-live, not just downstream takedown.
Deadlines & Compliance
- June 30, 2026: First FTC-mandated compliance reporting under Cox Media-style orders begins for "AI-powered" ad-tech vendors with active monitorships.
- August 2, 2026: EU AI Act GPAI enforcement (RFIs, model access, recalls) goes live one year after the obligations themselves; high-risk Annex III obligations now deferred to December 2, 2027 under the Omnibus deal.
- December 2, 2026: Shortened (3-month) grace period ends for AI-generated content transparency/labeling under the EU AI Act; nudifier and CSAM-generator prohibition takes effect the same day.
- Summer 2026: NIST expected to publish the Cybersecurity Framework Profile for AI plus predictive-AI control overlays; finalization slated for 2027.
Worth Reading
- What the EU AI Omnibus Deal Changes for the AI Act and What Lies Ahead — TechPolicy.Press's clearest mapping of which obligations slipped, which were added, and which transparency burdens actually tightened.
- EU AI Act Update: Timeline Relief, Targeted Simplification, and New Prohibitions — Covington's compliance-officer-grade timeline showing what high-risk providers can actually defer and what they cannot.
- Battle for AI Governance: The White House's Plan to Centralize AI Regulation and States' Continuous Opposition — Vorys's read on how the DOJ AI Litigation Task Force is likely to attack Colorado, California and TRAIGA in the coming term.
With the federal pen back in the drawer, the binding rules this quarter will be written by EU rapporteurs, state AGs, and a handful of federal magistrates — not the West Wing.