Artificial IntelligenceRegulations

News digest

Two layers. A live feed that pulls the latest AI-regulation headlines and refreshes on its own every six hours, and below it, original summaries of the moves that actually matter, each written by hand and linked to the source.

In depth

The moves that matter

September 4, 2026
The register

September briefing: one new state law, one regulation, one order, and a next deadline in 27 days

The register's monthly review. August was digestion after the August 2 double effective date, but the review turned up three things the register had wrong or missing: Connecticut's CART Act, signed in May and starting October 1, had been recorded as a stalled framework; the EU's Digital Omnibus has been a published regulation since July 27; and a June 2 executive order created the first federal frontier-model mechanism. All three are corrected on their pages. Next up: October 1 in Connecticut, December 2 in the EU, and the January 1 cluster.

United StatesHigh-risk systemsTransparency and disclosureSource ↗
August 2, 2026
European Commission

Enforcement begins: the AI Office takes over the AI Act

Beyond the transparency duties, August 2 was the day the AI Act grew teeth. The European Commission's AI Office, working with national authorities, formally began enforcing the act, and the Commission's power to fine providers of general-purpose AI models became applicable the same day. The obligations those fines attach to have been in force since August 2025; what changed is that noncompliance now has an enforcer with jurisdiction and a penalty schedule rather than a grace period.

European UnionGenerative and foundation modelsTransparency and disclosureSource ↗
July 15, 2026
CAC, NDRC, MIIT

China's agent era arrives as two rules, not one

Most coverage merges them; the register does not. The Implementation Opinions on Intelligent Agents, issued May 8 by the CAC, NDRC, and MIIT, are the first national framework anywhere to regulate agentic AI as its own category, requiring every agent's decisions to be tiered into human-only, approval-required, and autonomous before deployment, with filing and testing duties in sensitive sectors. The separate Interim Measures for anthropomorphic interactive services took binding effect July 15 and govern companion AI: security assessments, no engineered dependency, no virtual intimacy for minors. Major consumer products pulled features that day.

ChinaGenerative and foundation modelsTransparency and disclosureSource ↗
August 2, 2026
European Commission

The AI Act's transparency obligations take effect across the EU

Article 50 is now law in application: chatbots must disclose that they are machines, deepfakes and AI-written text on matters of public interest must be labeled, and emotion-recognition and biometric-categorization deployments must be disclosed to the people exposed to them. The one accommodation from the Digital Omnibus is narrow: generative systems already on the market before August 2 have until December 2, 2026 to meet the machine-readable marking duty, while systems launched from August 2 must mark immediately and deployer labeling duties carry no grace at all. Violations run to 15 million euros or 3 percent of worldwide turnover.

European UnionTransparency and disclosureDeepfakes and synthetic mediaSource ↗
August 2, 2026
Morgan Lewis

California's AI Transparency Act becomes operative the same day

The first wave of the AI Transparency Act (SB 942, as amended by AB 853) is now in force for covered providers: generative AI systems with more than one million monthly users publicly accessible in California must offer a free AI detection tool, apply visible disclosures, and embed latent, machine-readable disclosures in AI-generated image, video, and audio content. AB 853 aligned the date with the EU's deadline on purpose and queued expansion waves for large online platforms, hosting platforms, and capture devices through 2027 and 2028.

CaliforniaTransparency and disclosureDeepfakes and synthetic mediaSource ↗
July 24, 2026
Official Journal of the EU

The Digital Omnibus becomes law: Regulation (EU) 2026/1744

Published in the Official Journal on July 24 and in force from July 27, the amending regulation converts May's political agreement into binding text: Annex III high-risk duties move to December 2, 2027, Annex I product-integrated duties to August 2, 2028, generative systems already on the market get until December 2, 2026 for machine-readable marking, and a new prohibition on non-consensual intimate imagery generators lands the same December day. Nothing about Article 50 or general-purpose model duties was deferred.

European UnionHigh-risk systemsTransparency and disclosureSource ↗
July 20, 2026
European Commission

Commission adopts final Article 50 transparency guidelines, thirteen days before the deadline

The European Commission adopted the final 51-page guidelines on the AI Act's transparency obligations and assessed the Code of Practice on Transparency of AI-Generated Content as adequate. The final text softens the draft in one important place: image, audio, and video deepfakes generated before August 2, 2026 never need retroactive marking, with the date of generation governing, while AI-written text on matters of public interest is judged by its publication date, so older text published after the deadline must be labeled. Open-source systems get no exemption. Transparency violations carry fines up to 15 million euros or 3 percent of worldwide turnover.

European UnionTransparency and disclosureDeepfakes and synthetic mediaSource ↗
July 10, 2026
European Commission

EU AI Act transparency phase goes live; chatbot disclosure now enforceable

The AI Act's Article 50 transparency duties are now in effect. Any business deploying a chatbot or conversational system for EU users has to disclose that the user is dealing with AI, and providers of generative systems must mark synthetic output. Machine-readable marking for content from systems already on the market has a grace period to December 2, 2026.

European UnionTransparency and disclosureDeepfakes and synthetic mediaSource ↗
July 9, 2026
European Commission

Digital Omnibus receives final green light, locking in high-risk delay

The AI Act simplification package cleared its final approval. It confirms the deferral of stand-alone high-risk obligations to December 2, 2027 and product-embedded ones to August 2, 2028, narrows the 'safety component' test, and adds the nudifier and CSAM prohibition taking effect in December 2026.

European UnionHigh-risk systemsProhibited AI practicesSource ↗
July 7, 2026
European Commission

Commission presents EU Action Plan on Cybersecurity and AI

The Commission set out a coordinated plan to help member states and businesses handle the security risks of the most advanced AI models, and signaled a call to build EU capacity to evaluate models before they reach the market, expected operational by 2027.

European UnionGenerative AI and foundation modelsSource ↗
July 6, 2026
Skadden

Illinois signs the third frontier AI law, the first anywhere to mandate independent audits

Governor Pritzker signed the Artificial Intelligence Safety Measures Act (SB 315), putting Illinois beside California and New York with the same frontier template: models trained past 10^26 operations, developers above 500 million dollars in revenue, safety frameworks, transparency reports, and incident reporting. Illinois goes one step further than either coast: from January 1, 2028, large frontier developers must retain an independent third party to audit their compliance annually and publish the results. The act takes effect January 1, 2027, enforced exclusively by the attorney general, with no private right of action, and it preempts local AI regulation inside the state.

United States: IllinoisGenerative AI and foundation modelsHigh-risk systemsSource ↗
June 14, 2026
BD Emerson

UAE approves a federal Artificial Intelligence and Data Authority

The UAE consolidated its AI, data, and digital-government functions under a single federal authority, the clearest signal yet that its layered governance model is formalizing. The move follows the DIFC's AI-specific regulation reaching full enforcement in January and binding-in-practice central bank guidance for financial institutions in February.

United Arab EmiratesHigh-risk systemsSource ↗
June 10, 2026
European Commission

EU publishes Code of Practice on marking and labelling AI-generated content

The Commission published guidance on how providers can meet the Article 50 content-marking duties. It is voluntary, but it is the clearest signal yet on what regulators will expect from watermarking and provenance signals when the transparency rules bite.

European UnionTransparency and disclosureDeepfakes and synthetic mediaSource ↗
May 19, 2026
Travers Smith / European Commission

EU draft guidelines work through all eight high-risk categories

Draft Commission guidelines on high-risk classification walk through each Annex III area with examples. They clarify that intended purpose is judged from instructions, marketing, and documentation together, so a provider cannot dodge high-risk status by omitting a use from the manual.

European UnionHigh-risk systemsAutomated decisions and employmentSource ↗
May 14, 2026
Hunton

Colorado repeals its landmark AI Act and replaces it with a narrower ADMT law

Governor Polis signed a bill repealing the first comprehensive US state AI law before it ever took effect. The replacement drops the duty of care, risk programs, and impact assessments, and instead regulates automated decision-making technology through disclosures and consumer rights, effective January 1, 2027.

United States: ColoradoHigh-risk systemsAutomated decisions and employmentSource ↗
June 2, 2026
White House

Second AI executive order: voluntary 30-day pre-release vetting for frontier models

Signed June 2 after an earlier draft was pulled over competitiveness fears, the order asks agencies to design a voluntary framework under which developers of covered frontier models give the government up to 30 days of pre-release access for national-security and cyber vetting, creates an AI cybersecurity clearinghouse, and prioritizes criminal enforcement against AI-enabled attacks. No licensing, no preclearance, and it elevates the NSA and Treasury into central oversight roles for a White House that had championed a hands-off posture.

United StatesGenerative and foundation modelsSource ↗
May 27, 2026
Connecticut General Assembly

Connecticut enacts the CART Act, the broadest state AI law of 2026

After the 2025 version died under a veto threat, SB 5 passed the Senate 32 to 4 and was signed May 27 as Public Act 26-15. It is a 39-section omnibus that regulates named high-risk uses rather than imposing a general risk-management regime: automated employment decision tools, AI companions with minor protections, frontier-developer whistleblower rights, generative content provenance, AI-related layoff disclosures, and online safety for minors. Most provisions take effect October 1, 2026, companions January 1, 2027, employment-tool deployer duties October 1, 2027, enforced by the attorney general with no private right of action.

ConnecticutAutomated decisions and employmentTransparency and disclosureSource ↗
May 8, 2026
Covington

Commission publishes draft guidelines on the AI Act's Article 50 transparency duties

The European Commission's AI Office issued 40 pages of draft guidance interpreting the chatbot-disclosure, content-marking, and deepfake-labeling obligations that apply from August 2, 2026, with consultation closing June 3 and the final version expected before the deadline. The guidelines confirm no retrospective marking of content already published before August 2, and pair with the voluntary Code of Practice on Transparency covering the technical marking architecture. Transparency violations carry fines up to 15 million euros or 3 percent of worldwide turnover.

European UnionTransparency and disclosureDeepfakes and synthetic mediaSource ↗
May 7, 2026
Latham & Watkins

EU institutions agree the AI Act Omnibus, extending deadlines and adding prohibitions

Negotiators reached political agreement on the first amendments to the AI Act since adoption. The package extends high-risk deadlines, trims duplicative product rules, softens the AI-literacy duty, and introduces a ban on AI that generates non-consensual intimate imagery and CSAM.

European UnionHigh-risk systemsProhibited AI practicesSource ↗
March 11, 2026
Ropes & Gray

Analysis of Executive Order 14365 concludes the order cannot preempt state AI law on its own, since preemption generally flows from Congress. The FTC policy statement and Commerce evaluation of state laws were due this week, but courts will decide whether any of it actually displaces state rules.

United States (Federal)High-risk systemsTransparency and disclosureSource ↗
March 1, 2026
IAPP

Vietnam's AI Law takes effect, the first binding comprehensive statute in Southeast Asia

Enacted December 10, 2025 and effective March 1, Vietnam's standalone AI law brings risk classification, role-based obligations, and explicit extraterritorial reach to foreign providers serving Vietnamese users. It supersedes the AI chapter of the Digital Technology Industry Law within two months of that law taking effect, and the national registration portal opened July 1.

VietnamHigh-risk systemsTransparency and disclosureSource ↗
February 2026
The Reporter

Kenya tables an Artificial Intelligence Bill

Following its 2025 national AI strategy, Kenya's government tabled an AI bill proposing system classification, obligations for high-risk systems, a technology register, and transparency safeguards, one of the first live AI legislative processes in Africa.

KenyaHigh-risk systemsSource ↗
January 14, 2026
Focus Taiwan

Taiwan's AI Basic Act enters into force

Passed unanimously on December 23 and promulgated January 14, Taiwan's framework statute sets seven principles, names the National Science and Technology Council as competent authority, and starts a two-year clock for aligning sectoral laws. It imposes no private-sector obligations yet; the risk-classification framework from the digital affairs ministry is the piece to watch.

TaiwanTransparency and disclosureSource ↗
January 2026
CMS

Peru's AI regulation enters into force with prohibitions and phased sector deadlines

The implementing regulation of Law 31814 took effect ninety business days after its September publication, adding a three-tier risk framework, prohibitions on unauthorized mass surveillance and predictive policing, human-oversight duties, sandboxes, and sector compliance clocks running one to four years.

PeruProhibited AI practicesHigh-risk systemsSource ↗
January 1, 2026
King & Spalding

A wave of US state AI laws takes effect: California, Texas, Illinois

California's frontier transparency and training-data laws, Texas TRAIGA, and Illinois' AI employment amendment all took effect on the same day, a month after the federal preemption order. For now every one of them remains enforceable, and most follow the affected resident rather than the company's location.

United States: CaliforniaUnited States: TexasUnited States: IllinoisTransparency and disclosureGenerative AI and foundation modelsAutomated decisions and employmentSource ↗
December 11, 2025
The White House

Trump signs Executive Order 14365 to challenge and preempt state AI laws

The order creates a DOJ AI Litigation Task Force to challenge state AI laws, directs Commerce to flag 'onerous' laws, and conditions some federal funding on state compliance. It names Colorado's AI Act as an example and carves out state laws on child safety, data centers, and government procurement.

United States (Federal)High-risk systemsTransparency and disclosureSource ↗
December 2, 2025
ISMG

Australia shelves mandatory AI guardrails in its National AI Plan

Three months after consulting on ten mandatory guardrails for high-risk AI, Australia's government released a National AI Plan that relies on existing technology-neutral law and voluntary guidance instead, funding an AI Safety Institute to run gap analysis. The guardrails are shelved, not withdrawn.

AustraliaHigh-risk systemsSource ↗
September 1, 2025
China Law Translate

China's AI content-labeling rules take effect with dual visible and embedded marks

The CAC's labeling Measures now require explicit labels users can see and implicit labels inside file metadata on AI-generated content, backed by a mandatory national standard. Major platforms issued compliance statements as the rules went live.

ChinaTransparency and disclosureDeepfakes and synthetic mediaSource ↗
Sourcing

The watchlist

The in-depth summaries read official regulators and primary law first, then use specialist trackers to catch and confirm developments. The live feed draws current headlines from major outlets and links straight to the original.