DAILY BRIEFING · MONDAY, MAY 25, 2026
Brussels blinks on high-risk AI deadlines while Washington stalls its own preemption order — leaving U.S. states, the FTC, and corporate boards to fill an enforcement vacuum that is rapidly reshaping how AI systems must be governed, documented, and held to account.
⚡ QUICK TAKES
| Story | Signal |
|---|---|
| ↗ Council and Parliament agree to simplify and streamline AI rules | EU pushes high-risk AI deadlines back 16 months and tightens deepfake bans. |
| ↗ EU legislators agree to delay for high-risk AI rules | Delay is conditional on harmonized standards — deployers should not slow down work. |
| ↗ AI Act Update: EU Resolves to Change Rules and Extend Deadlines | Pro-industry recalibration — not a rollback; compliance baselines need a re-plan. |
| ↗ EU AI Act Update: Timeline Relief, Targeted Simplification, and New Prohibitions | Bias-detection data use is now expressly permitted under the Omnibus deal. |
| ↗ Proposed State Privacy and AI Law Update: May 25, 2026 | California chatbot bills move; second wave of state AI law is here. |
| ↗ White House postpones signing of AI executive order | Federal preemption push stalls; states keep the regulatory initiative. |
| ↗ White House 'studying' AI security executive order | FDA-style pre-release review under consideration for frontier models. |
| ↗ May 2026 Is the Forecast: AI Governance Just Became Data Governance | Regulators have folded AI governance into data-protection accountability. |
| ↗ Do you need a chief AI officer? How the tech is changing boardrooms | Board-level AI oversight lags adoption; CAIO role is solidifying. |
| ↗ UK government announces plans to set standards for how AI is deployed | UK shelves a standalone AI bill; sector regulators get the lead. |
| ↗ FTC can fine platforms $53,088 per deepfake left up after 48 hours | FTC deepfake enforcement turns 48-hour takedown into a legal SLA. |
| ↗ The TAKE IT DOWN Act is now law — what platform operators must have in place | Mid-sized platforms carry the heaviest TAKE IT DOWN compliance burden. |
| ↗ AI in litigation series: An update on AI copyright cases in 2026 | Fair-use line is hardening: training OK, piracy not. |
| ↗ Supreme Court declines to consider whether AI alone can create copyrighted works | Human authorship survives; the line on AI-assisted works stays unclear. |
| ↗ State Children's Online Safety Laws Expand to AI Chatbots | Chatbots are becoming regulated consumer products for minors. |
| ↗ OpenAI, Meta face Senate-backed AI child safety bill targeting chatbot access | Bipartisan federal bill would ban AI companions for minors. |
| ↗ The Agentic AI Governance Gap of Early 2026 | Most organisations can monitor agents but can't stop them. |
| ↗ AI and workplace discrimination after the EEOC and DOL rollbacks | Federal retreat shifts AI-hiring liability to private suits and state law. |
| ↗ FDA announces sweeping changes to oversight of wearables, AI-enabled devices | FDA narrows AI device perimeter; clinician review becomes the safety check. |
| ↗ Inside the AI Index: 12 takeaways from the 2026 report | Capability outruns governance; U.S. lead over China has nearly disappeared. |
| ↗ How cognitive manipulation and AI will shape disinformation in 2026 | Policy focus shifts from AI-content labels to provenance verification. |
Council of the EU · May 2026
EU co-legislators reached a provisional Omnibus VII deal on May 7 that postpones use-based high-risk obligations from August 2026 to December 2027 and product-regulated high-risk rules to August 2028. The agreement also introduces new prohibitions on non-consensual intimate AI imagery and extends SME-style relief to small mid-caps. It is the most consequential rewrite of the AI Act since its 2024 adoption.
✍️ Council of the EU Press Release · Read article →
Hogan Lovells · May 2026
Hogan Lovells walks through the Omnibus political deal, flagging that compliance for Annex III systems now hinges on harmonized standards being available — a 'standards-readiness' trigger that gives the Commission more control over the de facto start date. The firm warns deployers not to treat the delay as a holiday: documentation, FRIA and AI literacy obligations still bite.
✍️ Hogan Lovells AI Practice · Read article →
Latham & Watkins · May 2026
Latham parses the Omnibus changes alongside reinforced AI Office powers and a narrowed governance perimeter. The note highlights the December 2026 effective date for new prohibited practices and reads the deal as a modest pro-industry recalibration rather than a structural rollback. It maps out what existing AI Act compliance programmes need to re-baseline now.
✍️ Latham & Watkins LLP · Read article →
Covington — Inside Privacy · May 2026
Covington focuses on the practical operational impact: a permitted bias-detection use of special-category personal data, simplified registration in the EU database, and stricter rules on AI-generated intimate content. The piece argues the simplification track is meant to land before the August 2026 cliff and warns of fragmentation if Member States interpret the new flexibilities unevenly.
✍️ Covington & Burling LLP · Read article →
Troutman Pepper Locke · May 25, 2026
The weekly tracker logs a surge of California chatbot bills moving through Appropriations — AB 1988, SB 867, AB 2023, SB 1119 and AB 1609 — alongside amendments to the state's AI Transparency Act and movement on New York's S 1169 algorithmic discrimination bill. The pace, with seven state AI laws already on the books in 2025, signals a second wave of state-level rulemaking far more granular than the first.
✍️ Troutman Pepper Locke · Read article →
Nextgov / FCW · May 2026
An anticipated executive order targeting state AI laws for federal preemption was pulled from the President's signing schedule, with officials citing the need for further interagency review. The delay matters because the December 2025 order had already stood up a DOJ AI Litigation Task Force to challenge state laws, and industry expected a sharper preemption signal in May.
✍️ Nextgov / FCW · Read article →
Federal News Network · May 2026
A senior White House official told reporters the administration is weighing an FDA-style pre-release safety review for frontier models, with mandatory reporting tied to compute thresholds. The framing is the strongest signal yet that the administration is willing to layer federal safety oversight even as it tries to clear state law from the field.
✍️ Federal News Network · Read article →
Cybersecurity Insiders · May 2026
Drawing on UK ICO guidance and EDPB Opinion 28/2024, the piece argues regulators have collapsed AI governance into existing data-protection accountability structures. A striking data point: 78% of organisations cannot validate training data provenance — the exact evidence regulators now demand. The takeaway is that AI compliance is now a data lineage problem.
✍️ Cybersecurity Insiders · Read article →
CNBC · May 11, 2026
CNBC surveys the institutionalisation of AI at board level, reporting that 54% of directors still don't have AI as a standing agenda item even as 100% of organisations have AI on their 2026 roadmap. The piece tracks the rise of the Chief AI Officer role and debates whether AI risk belongs inside the audit committee or in a dedicated technology committee.
✍️ CNBC · Read article →
Bird & Bird · April 2026
Bird & Bird unpacks Tech Minister Liz Kendall's April 28 speech, which formalises the UK's decision not to introduce a standalone AI bill and instead route oversight through sector regulators under a forthcoming Regulating for Growth Bill. The approach widens the gap with the EU and bets the UK can lead on standards and sandboxes rather than statute.
✍️ Bird & Bird LLP · Read article →
Tech Times · May 19, 2026
The TAKE IT DOWN Act's platform compliance window opened May 19, giving the FTC authority to levy per-violation civil penalties of $53,088 for failures to remove non-consensual intimate AI imagery within 48 hours. The federal regime supersedes a patchwork of state laws and turns deepfake takedown into an enforceable trust-and-safety SLA — with the first criminal conviction under the Act already on the books.
✍️ Tech Times · Read article →
ComplianceHub.wiki · May 2026
A detailed compliance walk-through of the May 19 deadline, covering required notice-and-takedown workflows, hash-matching for known copies, and FTC investigative posture. The piece argues mid-sized platforms are the most exposed because they lack the trust-and-safety scale of hyperscalers but still meet the Act's coverage thresholds.
✍️ ComplianceHub · Read article →
Norton Rose Fulbright · May 2026
Norton Rose tracks where the major AI training cases stand after the $1.5B Bartz v. Anthropic settlement and the partial Kadrey v. Meta ruling — both finding training on lawfully obtained books is fair use, but storing pirated copies is not. The piece reads 2026 as the year fair-use boundaries crystallise for generative AI, with implications for licensing markets.
✍️ Norton Rose Fulbright · Read article →
Morgan Lewis · March 2026
By denying cert in Thaler v. Perlmutter, the Supreme Court left intact the rule that human authorship is required for U.S. copyright — a decision now reverberating through generative-AI product design. Morgan Lewis flags that the open question is not whether AI works are protectable, but how much human creative contribution is enough.
✍️ Morgan Lewis LLP · Read article →
MultiState · April 30, 2026
MultiState charts a wave of state laws — California's SB 243, New York's S-3008C, and bills moving in Idaho, Oregon, Washington, Maine and Nebraska — that require companion chatbots to detect self-harm signals, disclaim non-human status, and avoid sexual content with minors. The trend recasts AI chatbots as regulated consumer products for under-18 users.
✍️ MultiState Insider · Read article →
Bloomberg · April 30, 2026
A bipartisan Senate Judiciary measure — endorsed unanimously in committee — would require strict age verification, ban AI companions for minors, and prohibit chatbots from sending sexually explicit content or self-harm encouragement. Reporting frames it as the most aggressive federal action on minors and AI to date and a likely floor vote in 2026.
✍️ Bloomberg · Read article →
Lumenova AI · Early 2026
Lumenova quantifies the autonomy problem: 100% of organisations have agentic AI on their 2026 roadmap, but 63% cannot enforce purpose limitations on AI agents and 60% cannot quickly terminate a misbehaving one. The piece argues existing governance frameworks weren't designed for systems that act, and calls for kill-switches, scoped credentials, and continuous behavioural audits.
✍️ Lumenova AI · Read article →
Husch Blackwell · May 2026
Husch Blackwell argues the rescission of EEOC AI guidance and a White House executive order curtailing disparate-impact enforcement do not eliminate Title VII or ADA exposure — they just shift risk to private litigation and state regulators like Colorado and Illinois. Employers should expect more plaintiff-led litigation and tighter state audit duties.
✍️ Husch Blackwell LLP · Read article →
STAT News · January 2026
FDA guidance narrows oversight of clinician-reviewable AI software, leaving high-risk diagnostic and therapeutic tools fully regulated but treating many wellness and decision-support tools as outside the medical-device perimeter. Critics argue the move offloads risk to clinicians; the agency frames it as alignment with international ISO 13485 expectations.
✍️ STAT News · Read article →
Stanford HAI · April 2026
Stanford's flagship AI Index finds generative AI hit 53% population adoption in three years — faster than the PC or the internet — and that U.S. and Chinese frontier models now trade places at the top of leaderboards. The report's policy chapter argues governance frameworks are not keeping pace with capability, a finding being cited across this week's regulatory debates.
✍️ Stanford HAI · Read article →
World Economic Forum · March 2026
WEF's analysis pairs the Global Risks Report's top-ranked short-term threat — mis- and disinformation — with new findings on AI-driven cognitive manipulation. Among the data points: 58% of U.S. adults expect synthetic election lies to escalate. The piece argues policy must move from labelling synthetic content to verifying provenance.
✍️ World Economic Forum · Read article →