AI Compliance Impact Map — Week of May 15–22, 2026

AI Compliance Impact Map — Week of May 15–22, 2026

This week's global AI compliance briefing spans 17 confirmed regulatory developments across US federal, US state, EU/UK, and China. The headline is Trump's last-minute cancellation of a seven-page AI executive order (David Sacks intervention, May 21) — leaving US federal AI oversight undefined — while California signed the first-ever state AI workforce executive order, the FTC activated TAKE IT DOWN Act enforcement at $53,088/violation, and Connecticut's AIRT Act sits awaiting a governor signature with a ~May 29 deadline. The EU published draft high-risk AI classification guidelines with three material interpretive clarifications, and China's Hangzhou court established a landmark precedent ruling AI-driven dismissals unlawful under the Labor Contract Law.

Global AI Regulation & Compliance Map
Global AI Regulation & Compliance Map@NeoDrop Official
2026/5/23 · 6:28
購読 1 件 · コンテンツ 5 件
Three immediate priorities this week:
  1. HIGH — May 29 deadline: Connecticut SB5 (AIRT Act) — Governor Lamont has until approximately May 29 to sign or veto. If signed, it becomes the most comprehensive state AI law in the US, with developer obligations taking effect October 1, 2026.
  2. HIGH — TIDA enforcement is live as of May 19: FTC began enforcing the TAKE IT DOWN Act (TIDA) with civil penalties of $53,088 per violation. If your platform hosts user-generated content, a 48-hour NCII removal mechanism must be in place now.
  3. HIGH — EU Article 50 consultation closes June 3: Draft guidelines on AI transparency obligations (deepfake disclosure, chatbot identification, AI-generated content labeling) are open for comment only until June 3. The underlying obligations apply August 2, 2026.

US federal: the AI EO that wasn't

Trump cancels AI executive order signing at the last minute

The week's most significant federal development was what did not happen. President Trump canceled a planned signing ceremony for a seven-page AI executive order just hours before it was scheduled to take place in the Oval Office on May 21. 1
The draft order — obtained by POLITICO — would have established a voluntary 90-day pre-release review window for frontier AI models, during which developers could submit models to federal agencies before public launch. 2 The NSA would have been authorized to conduct classified assessments; Treasury would have set up information-sharing agreements between AI companies and critical-infrastructure cyber defenders; and the DOJ would have enforced the Computer Fraud and Abuse Act (CFAA) against AI-enabled unauthorized access. A notable self-limiting clause: "Nothing in this section shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement." 2
The cancellation was triggered by former AI Czar David Sacks, who called Trump directly on Thursday morning — without notifying his own staff — to raise concerns that a voluntary review framework could become mandatory over time and would slow AI development relative to China. 3 Trump said he "didn't like certain aspects" of the order and was unwilling to do anything that could "get in the way" of the US lead over China. 4 The White House has not announced a revised timeline for reintroduction.
Industry officials had already been pushing to shorten the review window from 90 days to 14 days and to shift oversight to the intelligence community rather than civilian agencies. Notably, the Federal AI Standards and Innovation Center (formerly CAISI) had already signed pre-deployment evaluation agreements with Google DeepMind, Microsoft, and xAI — alongside existing agreements with Anthropic and OpenAI — before the cancellation. 5
Compliance signal: No new federal AI obligation was created this week. The cancellation leaves federal AI oversight policy undefined. Companies engaged in voluntary model-evaluation programs with CAISI should continue those independently; they carry no legal mandate either way.
リンクプレビューを読み込んでいます…

TIDA enforcement is live: 15 platforms warned, 12 nudify sites put on notice

The Federal Trade Commission (FTC) began enforcing Section 3 of the TAKE IT DOWN Act (TIDA) on May 19, 2026 — one year after President Trump signed the law. 6 Civil penalties: $53,088 per violation.
TIDA requires covered platforms to provide a mechanism for victims to request removal of non-consensual intimate imagery (NCII) and to act on valid requests within 48 hours, removing the content and known identical copies. FTC Chair Andrew Ferguson sent stakeholder letters to 15 major platforms — including Alphabet, Amazon, Apple, Automattic, Bumble, Discord, Match Group, Meta, Microsoft, Snapchat, TikTok, and X — reminding them of their obligations. 7 On May 21, the FTC issued separate warning letters to 12 unnamed websites offering AI "nudify" tools — services that digitally strip clothing from uploaded photos. Ferguson stated: "Platforms no longer have any excuses — they must comply with their obligations under the TAKE IT DOWN Act or face the consequences." 7 The FTC also launched takeitdown.ftc.gov for victim complaints.
Compliance signal: Any platform hosting user-generated content that could include intimate imagery needs a live NCII removal request mechanism today. The 48-hour clock runs from receipt of a valid request.

Other federal items

H.R.8881 (SBA AI Utilization Act of 2026): Rep. Brad Finstad (R-MN-1) introduced a bill requiring the Small Business Administration (SBA) to submit annual reports on its internal AI use. The House Small Business Committee voted 23–0 to advance it on May 20. 8 No compliance implications for private entities.
White House FinTech EO (May 19): The White House signed "Integrating Financial Technology Innovation into Regulatory Frameworks," directing federal financial regulators to review barriers to fintech innovation within 90–180 days. The order does not contain AI-specific provisions or obligations. 9 Initial reports linking it to AI governance were incorrect.
Bartz v. Anthropic — settlement pending: The fairness hearing for the $1.5 billion Bartz v. Anthropic copyright class action settlement was held May 14 before Judge Araceli Martinez-Olguin in the Northern District of California. According to an observer analysis published May 22, the hearing went smoothly and the settlement appears likely to be approved, though the judge's formal order has not yet issued as of this writing. 10 The settlement covers 447,576 works at approximately $3,100 per work; class definition requires ISBN or ASIN and US Copyright Office registration. Separately, Elsevier v. Meta (filed May 5) adopted a narrower class definition limited to ISBN-registered works only — excluding ASIN/self-published titles — a restriction that may set a precedent in subsequent AI training data litigation.

US states

California: Newsom signs first-in-nation AI workforce executive order

Governor Gavin Newsom signed Executive Order N-6-26 on May 21, making California the first US state to issue an executive order specifically addressing AI's economic impact on workers and small businesses. 11
The order does not impose direct obligations on AI companies. It is a study-and-recommend directive, and explicitly states it creates no legally enforceable rights. 12 Key timelines for state agencies:
  • 90 days: Employment Development Department (EDD) launches a public dashboard showing AI's sector-by-sector employment impact, based on unemployment insurance data.
  • 180 days: Labor and Workforce Development Agency (LWDA) submits recommendations on updating California's WARN Act to provide earlier workforce disruption warnings, plus reviews of severance, work-share, and transitional employment programs.
  • October 15, 2026: Government Operations Agency recommends changes to incentive structures for public-interest AI development.
Newsom framed the signing as proactive: "California has never sat back and watched as the future happened to us — and we won't start now." 11 California Labor Federation President Lorena Gonzalez took a harder line: "Catastrophic job loss from AI is not inevitable, it's a political choice." 13 The order was signed one day after Meta announced layoffs of 8,000 employees.
Compliance signal: No direct employer obligation yet — this is the research phase. Monitor LWDA outputs; WARN Act amendments in California would directly affect any employer planning AI-driven workforce reductions.
リンクプレビューを読み込んでいます…

Connecticut SB5 (AIRT Act): signature clock is ticking

Connecticut's legislature transmitted Senate Bill 5 — the Artificial Intelligence Responsibility and Transparency Act (AIRT Act), designated Public Act 26-15 — to Governor Ned Lamont on May 14. 14 The legislature adjourned on May 6; Lamont has 15 days from receipt to act. Signature deadline: approximately May 29. All sources reviewed indicate Lamont has stated he intends to sign it. 15
SB5 passed 32–4 in the Senate and 131–17 in the House — the broadest legislative consensus of any comprehensive state AI bill to date. At 67+ pages, it is currently the most detailed US state AI law. Core provisions and effective dates:
AreaKey obligationEffective date
Automated employment decisions (AEDT)Developer documentation obligationsOct 1, 2026
AEDT deployerPre-decision consumer noticeOct 1, 2027
AI companion chatbotsDisclosure + prohibitions on romantic content / emotional manipulation with minorsJan 1, 2027
Synthetic digital contentC2PA-aligned provenance data for platforms with 1M+ monthly usersOct 1, 2026
Frontier model whistleblowersEmployee protections for safety reportingOct 1, 2026
AI regulatory sandboxIndependent third-party verification pilot (max 5 firms; sunsets 2030)Jul 1, 2027
Compliance signal: If Lamont signs, any company developing or deploying automated employment decision tools, AI chatbots for consumers, or synthetic content platforms faces October 1, 2026 obligations — under five months away. Gap assessment against SB5's developer documentation requirements should start now.

Colorado SB 26-189: law firm analysis wave arrives

Three weeks after Governor Jared Polis signed SB 26-189 on May 14 — replacing the 2024 Colorado AI Act (SB 24-205) with a narrower Automated Decision-Making Technology (ADMT) transparency framework — Norton Rose Fulbright, Mayer Brown, and Buchalter all published detailed compliance analyses this week. 16 17 18
The law regulates ADMT used in seven consequential-decision categories: education, employment (hiring, promotion, compensation, scheduling), residential real estate, financial and lending services, insurance, healthcare, and essential government services. Explicitly excluded: anti-malware, calculators, databases, firewalls, spellcheckers, general-purpose AI chatbots with acceptable-use policies, and tools used purely for summarizing, translating, or drafting. The law takes effect January 1, 2027, with enforcement exclusively by the Colorado Attorney General (no private right of action) and a 60-day cure period through January 1, 2030.
Two practitioner observations stand out. Norton Rose Fulbright noted that Colorado's ADMT definition is broader than California CCPA's ADMT regulations — Colorado covers "assistive" decision-making while California limits scope to systems that "replace or substantially replace" a human decision. Both regimes share a January 1, 2027 effective date, meaning companies operating in both states need to reconcile two frameworks simultaneously. 16 Buchalter warned: "The enforcement stay creates a window of opportunity, not a reason for delay" — the xAI litigation that paused SB 24-205 does not stay SB 26-189, and existing anti-discrimination statutes under Colorado and federal law remain fully in force regardless. 18

Other state items

Missouri SB 1019: The Missouri legislature passed SB 1019 on May 15, prohibiting AI therapy chatbots — specifically services advertised as providing therapy, mental health diagnosis, or representation by a mental health professional. First violation: $10,000; subsequent violations: $20,000. Awaiting governor signature. 15
Illinois SB 3114 (Downcoding Transparency Act): Passed the Illinois Senate 59–0 on May 14. Prohibits health insurers from using algorithms alone to downcode medical claims; all downcoding decisions must be made or reviewed by a human against AMA guidelines. 15 Now in the House Insurance Committee.
California AB 1988 (PAUSE Act — Preventing AI User Self Endangerment Act): Passed the California Assembly on May 21 and moved to the Senate. The bill addresses AI chatbot safety. 15 Details of chatbot-specific obligations are under Senate committee review.
Maryland HB 895 (Protection From Predatory Pricing Act): Signed by Governor Wes Moore on April 28; effective October 1, 2026. Prohibits large food retailers (15,000+ sq ft) and third-party food delivery platforms from using dynamic pricing based on individual consumer data. Any other business using algorithms or personal data to set individualized advertising prices must disclose: "This price is set by an algorithm or using your personal data." 19 Ballard Spahr assessed the law as a potential template for future algorithmic pricing regulation in financial services — with implications for ECOA, FHA, and state UDAAP frameworks even before any legislative expansion.

EU/UK

EC publishes draft guidelines on high-risk AI classification

The European Commission published draft guidelines on classifying high-risk AI systems under Article 6(5) of the EU AI Act on May 19. 20 The guidelines — originally due in February 2026 — cover three areas: (1) general classification principles under Annex I (AI as a product safety component) and Annex III (eight use-case categories); (2) interpretation of the Article 6(3) "procedural tasks" exception; and (3) how to handle complex and agentic systems.
Three clarifications carry immediate compliance weight:
  • Systems must be assessed as a whole: Decomposing a complex AI pipeline into sub-modules to avoid high-risk classification is not permitted. The system's combined functionality determines its classification. 21
  • Contractual exclusions do not override marketing: If a provider's documentation, marketing materials, or examples effectively promote high-risk uses, simply stating in terms of service that such uses are excluded is insufficient. The Commission's draft states: "Merely asserting (for example in the terms of service) that high-risk uses are excluded is insufficient to avoid the system from being considered high-risk, where the provider's overall presentation, examples, or product positioning effectively provides for or promotes such uses." 21
  • The procedural-task exception is narrower than anticipated: Article 6(3) allows providers to argue a system only performs preparatory work, not the actual consequential decision. The guidelines indicate this exception does not apply where the system's output involves profiling. 22
EU high-risk AI classification guidelines concept
EU Commission concept illustration for AI Act high-risk classification guidance 20
The guidelines are non-binding; final legal authority rests with the Court of Justice of the EU (CJEU). Public consultation runs through June 23, 2026 via the AI Act Single Information Platform. Confirmed compliance deadlines: standalone Annex III high-risk systems — December 2, 2027; Annex I product-embedded systems — August 2, 2028.
Compliance signal: AI providers who have been arguing they fall within the Article 6(3) exception should re-examine that position, particularly any system whose output includes a risk score, ranking, or profiling element. Governance teams that start self-classification work now against this draft will have a significant preparation advantage over those waiting for the final version.

UK: AI code of practice regulations take effect; copyright response published

SI 2026/425 — the Data Protection Act 2018 (Artificial Intelligence and Automated Decision-Making Code of Practice) Regulations 2026 — entered into force on May 7. 23 The regulations require the Information Commissioner's Office (ICO) to produce a statutory code of practice governing how personal data may be processed in AI development and automated decision-making. Alongside this, the ICO is currently consulting on a draft automated decision-making (ADM) guidance document — consultation closes May 29, 2026.
On May 15, the UK government issued its formal response to the House of Lords Communications and Digital Committee report "AI, copyright and the creative industries." 23 Key commitments: a public consultation on non-consensual digital replicas in summer 2026; an AI labelling taskforce with a mid-term report due autumn 2026; and a Creative Content Exchange — a licensed content marketplace between rights holders and AI developers — piloting in summer 2026. The government stated it "will not introduce reforms to copyright law unless or until it is confident that they will meet our objectives for the economy and for UK citizens." 23
The May 14 King's Speech announced a Regulating for Growth Bill that would include an AI Growth Lab (cross-economy AI sandbox), but as of May 22 the bill had not been introduced in Parliament. 23 Nicola Cain, CEO of Handley Gill, summarized the resulting UK regulatory picture as "a patchwork of measures… being lined up with potentially significant implications for the regulation of AI in the UK in the short-medium term." 23
Compliance signal: UK-based AI developers and automated-decision-making deployers should respond to the ICO's ADM consultation by May 29. This guidance will become the statutory baseline once the ICO publishes the formal code of practice under SI 2026/425.

China

TC260-005: China publishes AI Ethics Safety Guidelines 1.0

China's National Technical Committee on Cybersecurity Standardization (TC260) published TC260-005 — AI Application Ethics Safety Guidelines 1.0 on May 19 at the China Cyberspace Civilization Conference in Nanning, with official distribution via the Cyberspace Administration of China (CAC) on May 22. 24 The standard is the first systematic AI application ethics safety standard in China.
The guidelines address obligations across three stages of the AI lifecycle: development (data sourcing, model design), service provision (deployment, user interaction), and end use. Core principles include human primacy, fairness and justice, protection of lawful rights, and promotion of social trust. The standard is non-binding, though it provides a practical benchmark against which CAC enforcement reviews of generative AI services are increasingly measured.

Hangzhou court rules AI-based dismissal illegal: landmark labor ruling

The Hangzhou Intermediate People's Court released a landmark AI-related labor dispute ruling in late April — designated as a typical case guiding lower courts — which received wide coverage in the week of May 15–22. 25 The facts: a 35-year-old quality control manager (identified as Zhou) at a Hangzhou fintech company was told his role could be entirely replaced by AI, offered a demotion with a 40% pay cut from RMB 25,000 to RMB 15,000 per month, and dismissed when he refused. Both the Yuhang District Court (first instance) and the Hangzhou Intermediate Court (second instance) found the termination unlawful.
The court's stated rationale: A company's decision to adopt AI is a business and technology choice, not a "material change in objective circumstances" under Article 40 of China's Labor Contract Law — the provision that allows employers to modify or terminate contracts when business conditions change beyond either party's control. Judge Shi Guoqiang reasoned: "The objective circumstances that constitute grounds for contract termination are things like corporate mergers and asset transfers. Dismissing an employee merely because the company introduced AI to cut costs does not qualify." 25 The company was ordered to pay approximately RMB 260,000 in compensation (calculated at 2N — two months per year of service).
The court simultaneously issued "Guidelines on Protecting Workers' Rights in the Age of Artificial Intelligence" for the Hangzhou jurisdiction, directing companies to prioritize retraining and redeployment before considering AI-related layoffs.
Compliance signal for China operations: Workforce restructuring linked to AI adoption requires documented evidence that: (a) prior retraining or role reassignment was genuinely offered; and (b) the proposed alternative terms were commercially reasonable. A 40% pay cut was found unreasonable in this case. This ruling is now a cited precedent for similar cases in Hangzhou and increasingly referenced in other jurisdictions.
リンクプレビューを読み込んでいます…

CAC enforcement campaigns ongoing

The "Clean Internet · Rectify AI Disorder" (清朗·整治 AI 应用乱象) campaign launched April 30 remains active through approximately late August 2026. 26 The two-phase campaign targets 14 categories of AI application violations, including:
  • Phase 1 (service compliance): unregistered generative AI services, inadequate safety filters, training data security failures, AI data poisoning (e.g. GEO manipulation), non-compliant synthetic content labeling, AI tools used for network intrusion or identity theft, unmanaged open-source model deployments.
  • Phase 2 (content integrity): AI-modified classic works ("digital sewage"), deepfake impersonation, AI-generated misinformation on public affairs, content harmful to minors, AI-powered astroturfing/bot networks, unauthorized nudity/occult AI services.
Separately, CAC Order No. 21 (AI Anthropomorphic Interaction Service Interim Measures) takes effect July 15, 2026 — 53 days from publication of this issue. No implementation guidelines have been issued yet. Services with 1 million+ registered users or 100,000+ monthly active users must submit a safety assessment to provincial-level CAC before that date.

Upcoming compliance deadlines — next 30 days

PriorityDateObligationJurisdictionScope
HIGHMay 28, 2026China CNCERT AI cybersecurity testing program: registration closes. 8 test scenarios covering AI attack defense, vulnerability mining, large model safety.ChinaAI/cybersecurity companies
HIGHMay 29, 2026UK ICO draft automated decision-making (ADM) guidance: consultation closes. Statutory code of practice under SI 2026/425.UKCompanies using ADM in recruitment, credit, insurance, or any regulated sector
HIGHMay 29, 2026Connecticut SB5 (AIRT Act): governor signature deadline (approx.). If signed, developer obligations under AEDT provisions and synthetic content provenance take effect Oct 1, 2026.Connecticut, USAEDT developers, synthetic content platforms (1M+ monthly users), frontier model developers
HIGHMay 29, 2026FDA RFI on AI-optimized early-phase clinical trials: comment period closes (Docket FDA-2026-N-4390).US (federal)Clinical stage biotech and pharma companies, CROs
HIGHJune 3, 2026EU AI Act Article 50 transparency guidelines: public consultation closes. Underlying obligations (chatbot disclosure, AI-generated content labeling, deepfake disclosure) apply August 2, 2026.EUAll AI system providers and deployers interacting with users
WATCHJune 23, 2026EC draft high-risk AI classification guidelines: consultation closes.EUAI system providers conducting self-assessment under Annex I or Annex III
WATCHJuly 15, 2026China CAC Order No. 21 (AI Anthropomorphic Interaction): effective date. Services with 1M+ registered users must submit provincial CAC safety assessment before this date.ChinaAI companion, emotional support, digital-persona service providers
WATCHAugust 2, 2026EU AI Act Article 50 transparency obligations take effect for all new AI systems in market.EUAll AI system providers
WATCHOctober 1, 2026Colorado SB 26-189 ADMT deployer record-keeping begins (3-year retention); Maryland HB 895 algorithmic pricing disclosure effective.Colorado, Maryland, USADMT deployers in CO; food retailers and delivery platforms in MD
Cover image: California Governor Newsom's official graphic for Executive Order N-6-26 (AI Workforce Executive Order), May 21, 2026. Image from Governor's Office of California.

参考ソース

  1. 1Trump Cancels Signing of Executive Order Granting Oversight of A.I. Models
  2. 2Read Trump's unsigned AI executive order
  3. 3Trump yanked AI order after David Sacks raised industry concerns
  4. 4Trump postpones AI executive order signing: 'I didn't like certain aspects'
  5. 5Trump postpones executive order focused on AI security
  6. 6FTC Begins Enforcing the TAKE IT DOWN Act
  7. 7US Federal Trade Commission warns 12 'nudify' websites to comply with Take It Down Act
  8. 8H.R.8881 — SBA Artificial Intelligence Utilization Act of 2026
  9. 9Integrating Financial Technology Innovation into Regulatory Frameworks
  10. 10Has the Anthropic Settlement Changed Everything?
  11. 11Governor Newsom signs first-of-its-kind executive order to prepare workers and businesses for potential AI disruption
  12. 12Executive Order N-6-26
  13. 13California governor orders officials to find ways to mitigate AI layoffs
  14. 14Connecticut General Assembly — SB 5 Bill Status
  15. 15AI Legislative Update: May 22, 2026
  16. 16Colorado enacts revised AI law
  17. 17Colorado Enacts New ADMT Law Replacing Colorado AI Act
  18. 18Colorado Rewrites Its AI Law: What Employers Must Know About SB 26-189
  19. 19Mortgage Banking Update — May 21, 2026
  20. 20Commission seeks feedback on the draft guidelines for the classification of high-risk artificial intelligence systems
  21. 21EU AI Act: draft high-risk classification guidelines
  22. 22Guidelines support providers with 'high-risk' AI self-assessments
  23. 23Ne(a)r Hor(i)zon — The Future of UK Artificial Intelligence Regulation
  24. 24《人工智能应用伦理安全指引1.0》发布
  25. 25「你的岗位被AI取代了」，杭州35岁男子收到通知傻眼！法院判决：公司违法，支付2N赔偿金
  26. 26中央网信办部署开展「清朗·整治AI应用乱象」专项行动
Global AI Regulation & Compliance Map
Global AI Regulation & Compliance Map

このコンテンツについて、さらに観点や背景を補足しましょう。

  • ログインするとコメントできます。