Regulation Tracker

Controversies & regulation across sectors

Coming up and recent changes

Coming up

26 October 2026: Colorado: public hearing on the Attorney General's draft rules; written comments close the same day

2 December 2026: EU: deadline for marking AI-generated content, for providers whose systems were already on the market

1 January 2027: Colorado: SB 26-189 takes effect, and the Attorney General's rules are due

1 January 2027: New York: the RAISE Act takes effect

1 July 2027: California: SB 947 (no discipline or firing by automated system alone) and SB 1119 (companion-chatbot safety) take effect

2 December 2027: EU: high-risk system obligations apply in most areas

1 January 2028: Connecticut: the last provisions of SB 5 take effect

1 January 2028: California: SB 813 (independent AI verification bodies) takes effect

2 August 2028: EU: high-risk obligations apply to AI built into regulated products

What changed

Checked 4 October 2026

California closed its 2026 session with nearly 30 AI measures signed, among them a ban on firing or disciplining a worker by automated system alone (SB 947), companion-chatbot safety rules (SB 1119) and independent AI verification bodies (SB 813). Two health care AI bills were vetoed.

Connecticut's SB 5 began to apply: its first provisions took effect on 1 October 2026.

Texas opened the Attorney General's complaint portal on 1 September 2026, the route by which its AI law is enforced.

Colorado's Attorney General filed draft rules and set a public hearing for 26 October 2026. Enforcement remains stayed.

Brazil's AI bill has still not been voted on in the Chamber of Deputies; it awaits a committee opinion.

Added: Executive Order 14365, which directs federal challenges to state AI laws.

Added: New York's RAISE Act for frontier AI models, effective 1 January 2027.

Added: Illinois HB 3773 on AI in employment, in force since 1 January 2026.

Added: China's rules for human-like AI companions, in force since 15 July 2026.

Corrected: Japan's AI Promotion Act passed in 2025, not 2024.

Contested issues (12)

Deepfakes & election misinformation

Generative AI enables cheap, convincing synthetic audio/video of public figures, fueling disinformation around elections and conflict reporting.

Risk level: Critical

Sectors: Mass Media / Platforms, Public Sector / Security

Who is affected: Voters, journalists, public officials, electoral bodies

Why it is contested: Synthetic media can deceive at scale and at near-zero marginal cost, while detection lags generation. Liability for platforms and publishers remains unsettled.

Examples: AI-generated robocalls and deepfake videos targeting voters in 2024 election cycles; fabricated images of conflict zones shared as real.

Related laws: EU AI Act — Prohibited practices; EU AI Act — Deployer transparency (Article 50); China — Deep Synthesis Provisions; Tennessee ELVIS Act; Brazil — AI Bill (PL 2338/2023); Connecticut AI Act (SB 5, 2026)

Sources: European Commission — AI Act framework; Library of Congress — China Deep Synthesis Provisions

Generative AI & copyright infringement

AI firms train on copyrighted works without consent or compensation; rights holders sue while models continue to launch.

Risk level: High

Sectors: Creative / IP, Mass Media / Platforms

Who is affected: Authors, publishers, artists, record labels, AI developers

Why it is contested: Turns on whether training is fair use or licensing-required, and whether outputs substantially resemble protected works. Outcome reshapes the economics of the creative industries.

Examples: The New York Times v. OpenAI; artist class actions against Stable Diffusion; music labels over AI training.

Related laws: EU AI Act — General-purpose AI (GPAI) rules; US Copyright Office — AI guidance; Tennessee ELVIS Act

Sources: Harvard Law — AI, copyright & the Drake case

AI voice cloning of artists

Voice-synthesis tools clone performers' voices without consent, threatening musicians and voice actors.

Risk level: High

Sectors: Creative / IP

Who is affected: Musicians, voice actors, labels, fans

Why it is contested: A voice is not traditionally protected like a likeness; cloning is now trivial and commercially damaging, outrunning common-law publicity rights.

Examples: Ghostwriter977's 'Heart on My Sleeve' mimicking Drake and The Weeknd, pulled from streaming platforms.

Related laws: Tennessee ELVIS Act; China — Deep Synthesis Provisions

Sources: Harvard Law — AI, copyright & the Drake case; Skadden — Tennessee ELVIS Act

AI bias in hiring & employment

Automated CV-screening and video-interview tools discriminate against protected groups at scale.

Risk level: High

Sectors: Commercial / Workplace

Who is affected: Job applicants, women, minorities, disabled workers

Why it is contested: Opaque scoring entrenches historical bias; applicants rarely know they were filtered by an algorithm, and auditing is technically difficult.

Examples: AI screening tools that down-rank women's CVs; the Mobley v. Workday class action over AI-driven hiring bias.

Related laws: NYC Local Law 144 (AEDT); Illinois AI Video Interview Act; Colorado AI Act (SB 24-205 → SB 26-189 ADMT); EU AI Act — High-risk systems (Article 26); South Korea — AI Basic Act; Connecticut AI Act (SB 5, 2026); Illinois AI in employment (HB 3773)

Sources: European Commission — high-risk AI use cases

AI-assisted cheating in education

ChatGPT and similar tools enable plagiarism and undermine assessment integrity across schools and universities.

Risk level: Medium

Sectors: Education

Who is affected: Students, educators, institutions, accreditation bodies

Why it is contested: Bans are unenforceable and detection tools produce false accusations; institutions split between prohibition and integration.

Examples: Mass ChatGPT adoption in 2023; flawed AI-detector tools flagging original student work as machine-generated.

Related laws: EU AI Act — High-risk systems (Article 26); US — Trump EO 'Promoting AI Development' & OMB memo; South Korea — AI Basic Act

Sources: European Commission — high-risk AI in education

AI & scientific research integrity

LLM hallucinations and AI-assisted authorship inject fabricated citations, data, and peer-review into the scientific record.

Risk level: High

Sectors: Scientific / Medical, Education

Who is affected: Researchers, reviewers, patients, publishers

Why it is contested: Hallucinated references and methods are hard to catch; AI-generated papers risk polluting evidence bases used in medicine and policy.

Examples: Retracted AI-authored papers; fabricated datasets passing initial peer review.

Related laws: US — Trump EO 'Promoting AI Development' & OMB memo; EU AI Act — High-risk systems (Article 26); US FDA — AI/ML Software as a Medical Device (SaMD)

Sources: European Commission — AI safety components & medical AI; FDA — Artificial Intelligence in Software as a Medical Device

AI surveillance & workplace monitoring

Algorithmic management and biometric monitoring track workers continuously, raising privacy and autonomy concerns.

Risk level: High

Sectors: Commercial / Workplace, Public Sector / Security, Finance / Credit

Who is affected: Employees, gig workers, migrants, detainees

Why it is contested: Continuous scoring and emotion-recognition in workplaces pressure workers and enable opaque discipline, often without meaningful consent.

Examples: Warehouse productivity algorithms; emotion-recognition in hiring; remote biometric identification in public spaces.

Related laws: EU AI Act — Prohibited practices; China — Algorithmic Recommendation Provisions; UK — Online Safety Act & pro-innovation framework

Sources: European Commission — prohibited AI practices

Open-weight AI & dual-use security

Powerful open-weight models lower the barrier to building cyber, biological, and disinformation tools.

Risk level: Critical

Sectors: Scientific / Medical, Public Sector / Security, Commercial / Workplace

Who is affected: General public, critical infrastructure, democratic institutions

Why it is contested: Open release accelerates innovation and decentralization but also diffusion of capability to malicious actors; governance is nascent and contested.

Examples: Debate over releasing frontier model weights; export-control tensions around advanced chips and open models.

Related laws: US — Trump EO 'Promoting AI Development' & OMB memo; EU AI Act — General-purpose AI (GPAI) rules; New York RAISE Act (frontier AI models)

Sources: European Commission — GPAI & systemic-risk rules

AI 'nudification' & non-consensual intimate imagery

Apps generate synthetic sexual imagery of real people, overwhelmingly targeting women and minors.

Risk level: Critical

Sectors: Personal / Consumer, Mass Media / Platforms

Who is affected: Women, minors, public figures, victims of abuse

Why it is contested: Causes severe harassment and extortion harms; enforcement against app operators and hosting platforms remains weak.

Examples: AI 'nudify' apps targeting minors and celebrities; deepfake intimate-image abuse.

Related laws: EU AI Act — Prohibited practices; China — Deep Synthesis Provisions; Tennessee ELVIS Act

Sources: European Commission — Prohibition 9 (CSAM & intimate content)

AI social scoring & manipulation

Systems that score, rank, or manipulate people's behavior based on opaque criteria threaten autonomy and equality.

Risk level: High

Sectors: Public Sector / Security, Personal / Consumer, Finance / Credit

Who is affected: Citizens, consumers, marginalized communities

Why it is contested: Social scoring can lock people out of credit, housing, and services with no recourse; manipulation at scale undermines free choice.

Examples: Concerns over scoring-style systems; behavioral manipulation in commercial apps.

Related laws: EU AI Act — Prohibited practices; China — Algorithmic Recommendation Provisions; Colorado AI Act (SB 24-205 → SB 26-189 ADMT); China — Rules for human-like AI companions

Sources: European Commission — prohibited AI practices

AI market concentration & compute

Frontier AI is dominated by a handful of firms with capital, data, and compute advantages, raising competition and access concerns.

Risk level: High

Sectors: Commercial / Workplace, Public Sector / Security

Who is affected: Startups, researchers, consumers, regulators

Why it is contested: Concentration may entrench incumbents, limit scrutiny, and concentrate dual-use capability; remedies (interoperability, openness) are contested.

Examples: Investment and partnership scrutiny; antitrust attention to cloud-AI bundling.

Related laws: US — Trump EO 'Promoting AI Development' & OMB memo; UK — Online Safety Act & pro-innovation framework; EU AI Act — General-purpose AI (GPAI) rules; US — Executive Order 14365 on state AI laws

Sources: European Commission — GPAI obligations

Agentic AI & autonomous-action liability

When an AI agent books, pays, or commits on a user's behalf, who is liable for unauthorized actions? Agency, contract, and product-liability law were not written for non-human autonomous actors.

Risk level: High

Sectors: Commercial / Workplace, Finance / Credit, Personal / Consumer, Public Sector / Security

Who is affected: SMEs deploying agents, consumers, vendors, payment networks, regulators

Why it is contested: An agent that takes real-world action (spend money, make a booking, bind a contract) creates a liability gap: the deployer didn't directly authorize the specific act, the vendor didn't control the context, and the 'agent' is not a legal person. Agency-law theories are being tested but remain unsettled.

Examples: AI agents booking travel or making purchases; customer-service agents making promises or refunds; trading agents executing unauthorized positions.

Related laws: EU AI Act — High-risk systems (Article 26); US — Trump EO 'Promoting AI Development' & OMB memo; Japan — Soft-law AI framework (non-binding); UK — Online Safety Act & pro-innovation framework; New York RAISE Act (frontier AI models)

Sources: IMDA Singapore — Legal Responsibility for AI Agents; AI Standard of Care — Agentic AI & Autonomous System Liability; Mayer Brown — Contract Issues in Agentic AI Deals

Laws and rules (26)

EU AI Act

The first comprehensive risk-based AI law. Classifies systems by risk tier (minimal, limited, high, prohibited) and phases in obligations through 2028.

Jurisdiction: EU

Status: Phased (in force)

Key date: 2024-08-01 (entry into force)

Issued by: European Commission / Parliament / Council

Instrument: Regulation (binding)

Scope: All AI systems placed on the EU market or whose output is used in the EU

Key obligations: Prohibited-practice bans, GPAI transparency, high-risk conformity and human oversight, transparency labelling of AI-generated content

If you use AI in your business: Most SMEs are deployers, not providers. Your duties arrive in waves: from Aug 2026, tell people when they're chatting with AI or viewing AI-generated content; from Dec 2027, if you deploy a high-risk system (e.g. CV screening), follow the provider's instructions, keep human oversight, log activity for 6 months, and inform affected workers.

Related issues: Deepfakes & election misinformation; AI bias in hiring & employment; AI-assisted cheating in education; AI surveillance & workplace monitoring; AI social scoring & manipulation; Agentic AI & autonomous-action liability

Sources: European Commission — AI Act regulatory framework

EU AI Act — Prohibited practices

Bans 9 categories of AI including social scoring, untargeted facial-recognition scraping, workplace emotion-recognition, and (from Dec 2026) non-consensual intimate-content generation.

Jurisdiction: EU

Status: In force

Key date: 2025-02-02 (Prohibitions 1–8); 2026-12 (Prohibition 9)

Issued by: European Commission / AI Office

Instrument: Regulation (binding)

Scope: Providers and deployers placing or using prohibited systems in the EU

Key obligations: Prohibition on placing/using listed AI practices; fines up to €35M or 7% global turnover

If you use AI in your business: Binding now (since Feb 2025). If you deploy any of the nine banned practices — social scoring, workplace emotion-recognition, untargeted face-scraping — you're already on the wrong side of the law, regardless of who built the system.

Related issues: AI social scoring & manipulation; AI surveillance & workplace monitoring; AI 'nudification' & non-consensual intimate imagery; Deepfakes & election misinformation

Sources: European Commission — prohibited AI practices

EU AI Act — General-purpose AI (GPAI) rules

Obligations on GPAI model providers for transparency, copyright, and systemic-risk assessment; GPAI Code of Practice published July 2025.

Jurisdiction: EU

Status: In force

Key date: 2025-08-02

Issued by: European Commission / AI Office

Instrument: Regulation (binding) + voluntary Code

Scope: Providers of GPAI models placed on the EU market

Key obligations: Technical documentation, training-content summary, copyright policy, systemic-risk assessment & mitigation for frontier models

If you use AI in your business: Binds the model developer (e.g. OpenAI, Anthropic), not the SME calling the API. But if you build a high-risk product on a GPAI model, you inherit provider duties for that product.

Related issues: Generative AI & copyright infringement; Open-weight AI & dual-use security; AI market concentration & compute

Sources: European Commission — Guidelines for GPAI providers; EU AI Act — Article 51 (systemic-risk classification)

EU AI Act — Deployer transparency (Article 50)

From August 2026, deployers of chatbots, emotion-recognition, and generative-AI tools must inform people they are interacting with AI, and providers must label AI-generated content including deepfakes.

Jurisdiction: EU

Status: Phased (in force)

Key date: 2026-08-02; 2026-12-02 (content marking by providers of systems already on the market)

Issued by: European Commission / AI Office

Instrument: Regulation (binding)

Scope: Generative-AI providers and deployers of chatbots / emotion-recognition / AI content

Key obligations: Notify users of AI interaction; label AI-generated content; mark deepfakes on public-interest matters

If you use AI in your business: The SME chatbot rule — in force since Aug 2026. If you run a chatbot or deploy emotion-recognition, you must tell users they're interacting with AI. This is a deployer duty, not just the model provider's problem.

Related issues: Deepfakes & election misinformation; Generative AI & copyright infringement; AI surveillance & workplace monitoring

Sources: European Commission — AI Act regulatory framework; EU AI Act — Article 50 (deployer transparency); Jones Walker: the EU approved a high-risk delay, but most transparency obligations remain

EU AI Act — High-risk systems (Article 26)

From December 2027, high-risk AI (biometrics, critical infrastructure, education, employment, migration, justice) faces conformity, data-governance, human-oversight and logging duties — and deployers carry their own Article 26 obligations.

Jurisdiction: EU

Status: Phased (upcoming)

Key date: 2027-12-02 (most areas); 2028-08-02 (Annex I products)

Issued by: European Commission / Member State authorities

Instrument: Regulation (binding)

Scope: Providers and deployers of high-risk AI systems

Key obligations: Provider: risk management, quality datasets, logging, documentation, conformity. Deployer (Art 26): follow instructions, ensure human oversight, keep logs 6 months, inform affected persons

If you use AI in your business: If you deploy a high-risk system (hiring, credit, education), Article 26 binds you from Dec 2027: follow the provider's instructions, ensure competent human oversight, keep logs for 6 months, and notify workers' representatives and affected individuals.

Related issues: AI bias in hiring & employment; AI-assisted cheating in education; AI & scientific research integrity; AI surveillance & workplace monitoring; Agentic AI & autonomous-action liability

Sources: EU AI Act — Article 26 (deployer obligations); European Commission — high-risk AI use cases

US — Biden Executive Order 14110

Required frontier-model developers to report safety test results and set federal-agency AI governance via OMB memos. Revoked by Trump in January 2025.

Jurisdiction: US (federal)

Status: Revoked

Key date: Issued 2023-10-30; revoked 2025-01-20

Issued by: White House / OMB

Instrument: Executive Order + OMB Memoranda

Scope: Federal agencies and frontier-model developers

Key obligations: Safety-test reporting, agency AI governance, acquisition guidance

If you use AI in your business: Bound frontier developers and federal agencies, not ordinary SME deployers. Revoked Jan 2025 and replaced by the Trump EO.

Related issues: Open-weight AI & dual-use security; AI & scientific research integrity; AI market concentration & compute

Sources: Wiley Law — Trump revokes Biden AI EO

US — Trump EO 'Promoting AI Development' & OMB memo

Replaced the Biden framework with a pro-innovation posture; new OMB guidance on federal AI use and procurement issued in 2025.

Jurisdiction: US (federal)

Status: In force

Key date: EO 2025-01-23; OMB memo 2025-04

Issued by: White House / OMB

Instrument: Executive Order + OMB Memorandum

Scope: Federal agencies; AI use and procurement

Key obligations: Federal AI-use governance, acquisition reform, innovation promotion

If you use AI in your business: Governs how federal agencies use and buy AI. Not directly binding on private SMEs, but shapes what AI the US government will procure — a market signal for vendors.

Related issues: Open-weight AI & dual-use security; AI market concentration & compute; AI & scientific research integrity; Agentic AI & autonomous-action liability

Sources: Wiley Law — Trump revokes Biden AI EO; Inside Government Contracts — Trump-era OMB AI requirements

US — Executive Order 14365 on state AI laws

'Ensuring a National Policy Framework for Artificial Intelligence' directs federal agencies to challenge state AI laws in court and through funding and agency action, and to draft a federal law that would override them. The order does not itself override any state law, and Congress has passed no such law.

Jurisdiction: US (federal)

Status: In force

Key date: Signed 2025-12-11

Issued by: White House / Department of Justice / Commerce / FCC / FTC

Instrument: Executive order

Scope: Federal agencies; aimed at state AI laws

Key obligations: Attorney General to set up an AI Litigation Task Force to challenge state AI laws; Commerce to tie some federal funding to states' AI rules; FCC to consider a federal reporting standard for AI models; FTC to state when federal law overrides state requirements; officials to draft a federal override law. Child safety, data center infrastructure and states' own use of AI are left out of the planned override

If you use AI in your business: Puts no duty on your business. It matters because it puts state AI laws you may be following, such as Colorado's, under federal challenge: the Justice Department joined the lawsuit against Colorado's law in April 2026. Until a court or Congress says otherwise, state laws still apply.

Related issues: AI market concentration & compute; AI bias in hiring & employment

Sources: Gibson Dunn: President Trump's latest executive order on AI seeks to preempt state laws; StateScoop: state AI law moratorium omitted from the 2026 defense bill

Colorado AI Act (SB 24-205 → SB 26-189 ADMT)

The first comprehensive US state AI law was repealed and replaced mid-2026. SB 26-189 (the Automated Decision-Making Technology Act) drops the original 'reasonable care against algorithmic discrimination' duty in favour of a deployer disclosure-and-explanation framework, effective 1 Jan 2027 — with enforcement stayed pending federal litigation.

Jurisdiction: US (Colorado)

Status: Enacted (pending)

Key date: SB 24-205 signed 2024-05-17; SB 26-189 signed 2026-05-14, effective 2027-01-01 (enforcement stayed); AG draft rules filed 2026-08-11, hearing 2026-10-26

Issued by: Colorado General Assembly / AG

Instrument: State statute

Scope: Developers and deployers of covered automated decision-making tools (ADMT) in Colorado

Key obligations: Developer: technical documentation to deployers. Deployer: conspicuous pre-decision notice; plain-language explanation within 30 days of an adverse outcome; retain compliance docs 3 years; meaningful human review where feasible

If you use AI in your business: Enforcement is currently stayed by a federal court (xAI v. Colorado AG, April 2026) and won't begin until the AG completes rulemaking. The AG filed draft rules on 11 August 2026 and holds a public hearing on 26 October 2026; written comments are accepted until that day. When it does, deployers of covered decision tools must give consumers conspicuous notice before a material decision and a plain-language explanation within 30 days of any adverse outcome. Note: the original 'reasonable care' anti-discrimination duty was dropped in the 2026 rewrite.

Related issues: AI bias in hiring & employment; AI social scoring & manipulation

Sources: Colorado General Assembly — SB 24-205; AI Compliance Atlas — Colorado AI Act & SB 26-189 status; Ogletree: Colorado seeks input on proposed AI rules ahead of rulemaking hearing

California AI laws (SB 1047 vetoed; 2026 package signed)

Governor Newsom vetoed the frontier-model safety bill SB 1047 (2024), then signed transparency, deepfake, training-data and algorithmic-pricing laws. The 2026 session added nearly 30 AI measures, closing on 30 September 2026 with worker protections against AI-only discipline (SB 947), companion-chatbot safety rules (SB 1119) and independent AI verification bodies (SB 813). Two labor-backed health care AI bills (AB 2575, SB 903) were vetoed.

Jurisdiction: US (California)

Status: Mixed

Key date: SB 1047 vetoed 2024-09-29; AB 325 pricing effective 2026-01-01; 2026 package signed 2026-09-09 to 2026-09-30; SB 947 and SB 1119 effective 2027-07-01; SB 813 effective 2028-01-01

Issued by: California Legislature / Governor

Instrument: State statutes

Scope: AI providers, large generative models, companion chatbots, employers using automated decision systems, election deepfakes, pricing algorithms

Key obligations: Training-data disclosure, provenance marking, deepfake election restrictions, common-pricing-algorithm limits. From 1 July 2027: a human must corroborate before an automated system disciplines or fires a worker, with notice afterward (SB 947); companion chatbots need risk assessments, crisis protocols and child-safe defaults (SB 1119)

If you use AI in your business: A patchwork: SB 1047 (frontier developer safety) was vetoed; transparency and training-data disclosure laws bind developers, deepfake election rules bind distributors, and AB 325 binds anyone using a common pricing algorithm. New for employers: from 1 July 2027, SB 947 bars relying on an automated system alone to discipline or fire a worker. A person must check the decision first, and the worker must be told afterward.

Related issues: Deepfakes & election misinformation; Open-weight AI & dual-use security; Generative AI & copyright infringement

Sources: European Commission — AI Act framework (comparative); Orrick — US State AI Law Tracker; Kelley Drye: California's 2026 legislative session wraps; Transparency Coalition: 11 more AI laws enacted, 30 September 2026; CalMatters: Newsom enacts only some AI regulations sought by unions

Texas Responsible AI Governance Act (TRAIGA, HB 149)

Prohibits specific harmful AI uses (manipulation, discrimination, unlawful deepfakes) via an intent-based liability framework; third US state to adopt a comprehensive AI law. In force since 1 Jan 2026.

Jurisdiction: US (Texas)

Status: In force

Key date: Signed 2025-06-22; effective 2026-01-01; AG complaint portal open 2026-09-01

Issued by: Texas Legislature / Governor

Instrument: State statute (HB 149)

Scope: Developers and deployers of AI in Texas

Key obligations: Ban on specified harmful AI practices; baseline developer/deployer duties; regulatory sandbox; state AI council; AG-exclusive enforcement

If you use AI in your business: In force since Jan 2026. Intent-based: binds developers and deployers who knowingly use AI to manipulate, discriminate, or create unlawful deepfakes. Ordinary commercial use is not targeted; the law reaches bad actors. Enforcement is exclusive to the Texas AG, whose online complaint portal opened on 1 September 2026. A complaint lets the AG demand records; the AG must give written notice and wait 60 days before suing.

Related issues: AI social scoring & manipulation; AI 'nudification' & non-consensual intimate imagery; Deepfakes & election misinformation

Sources: Latham & Watkins — Texas TRAIGA signed into law; DLA Piper — Texas adopts the Responsible AI Governance Act; Cloud Security Alliance: Texas's TRAIGA complaint portal

Connecticut AI Act (SB 5, 2026)

The most comprehensive state AI law of the 2026 session: developer-to-deployer transparency, chatbot controls, automated-decision disclosures, a regulatory sandbox, and a study of certified third-party auditors.

Jurisdiction: US (Connecticut)

Status: Phased (in force)

Key date: Signed 2026-05-27 (Public Act 26-15); first provisions in force 2026-10-01; phased through 2028-01-01

Issued by: Connecticut General Assembly

Instrument: State statute

Scope: Developers and deployers of AI in Connecticut

Key obligations: Developer disclosures to deployers; deployer chatbot controls & automated-decision transparency; sandbox; auditor study

If you use AI in your business: Brings the EU-style developer/deployer split to a US state. The first provisions took effect on 1 October 2026, including the rules for automated employment decisions and a duty to tell the Labor Department when a mass layoff is tied to AI; the rest phase in through 1 January 2028. If you deploy a chatbot or automated decision system in Connecticut, expect disclosure duties and chatbot controls; developers must hand deployers the information needed to use the system safely.

Related issues: AI bias in hiring & employment; Deepfakes & election misinformation; AI-assisted cheating in education

Sources: CDT — 2026 State and Federal AI Legislation Updates; Orrick — US State AI Law Tracker; WilmerHale: What companies should know about Connecticut's new omnibus AI law; Kilpatrick: Connecticut enacts new AI legislation (SB 5)

Tennessee ELVIS Act

First state law protecting voice and likeness from AI deepfakes; expands right-of-publicity to cover generative-AI voice cloning.

Jurisdiction: US (Tennessee)

Status: In force

Key date: Signed 2024-03-21

Issued by: Tennessee Legislature / Governor

Instrument: State statute

Scope: AI developers, platforms, and users in Tennessee

Key obligations: Liability for unauthorized use of a person's voice or likeness via AI

If you use AI in your business: Binds anyone who uses AI to clone a person's voice or likeness without consent — developers, platforms, and end-users alike.

Related issues: AI voice cloning of artists; Generative AI & copyright infringement; AI 'nudification' & non-consensual intimate imagery

Sources: Skadden — Tennessee ELVIS Act; Latham & Watkins — ELVIS Act (PDF)

NYC Local Law 144 (AEDT)

Requires bias audits and transparency for automated employment decision tools used to screen candidates or employees.

Jurisdiction: US (New York City)

Status: In force

Key date: Enforced from 2023-07-05

Issued by: NYC Department of Consumer & Worker Protection

Instrument: Municipal ordinance

Scope: Employers using automated hiring tools in NYC

Key obligations: Independent bias audit, public summary, candidate notice

If you use AI in your business: Purely a deployer (employer) rule. If you use an automated tool to screen NYC candidates or employees, you must commission an independent bias audit, publish a summary, and notify candidates.

Related issues: AI bias in hiring & employment; AI surveillance & workplace monitoring

Sources: European Commission — high-risk AI use cases (comparative)

New York RAISE Act (frontier AI models)

New York's Responsible AI Safety and Education Act puts safety and reporting duties on the developers of the largest AI models. Signed in December 2025 and finalized by an amendment in March 2026, which brought it closer to California's frontier-model transparency law.

Jurisdiction: US (New York)

Status: Enacted (pending)

Key date: Signed 2025-12-19; amendment signed 2026-03-27; effective 2027-01-01

Issued by: New York Legislature / Governor

Instrument: State statute

Scope: Developers of frontier models (trained with more than 10^26 operations); extra duties for those with annual revenue above $500 million

Key obligations: Publish a safety framework and update it yearly; publish a transparency report before releasing a new or changed model; report a critical safety incident within 72 hours; quarterly risk summaries to a state office. Enforced by the Attorney General, with penalties up to $1 million for a first violation and $3 million after; no private lawsuits

If you use AI in your business: Binds only the companies that train the very largest models. A business that uses those models has no duties under it.

Related issues: Open-weight AI & dual-use security; Agentic AI & autonomous-action liability

Sources: Wiley: New York finalizes RAISE Act for frontier AI models; Davis Wright Tremaine: NY overhauls frontier AI transparency law

Illinois AI Video Interview Act

Requires employer consent, disclosure, and data-deletion rights when AI analyzes video job interviews.

Jurisdiction: US (Illinois)

Status: In force

Key date: Effective 2020-01-01 (amended 2023)

Issued by: Illinois General Assembly

Instrument: State statute

Scope: Employers using AI to evaluate video interviews in Illinois

Key obligations: Notify applicants, obtain consent, share evaluation data, allow deletion

If you use AI in your business: Binds employers (deployers) using AI to evaluate video interviews in Illinois: notify applicants, get consent, share evaluation data, and allow deletion.

Related issues: AI bias in hiring & employment; AI surveillance & workplace monitoring

Sources: European Commission — high-risk AI use cases (comparative)

Illinois AI in employment (HB 3773)

Amends the Illinois Human Rights Act so that using AI in a way that discriminates in employment, or using it in employment decisions without telling people, is a civil rights violation. In force since January 2026, though the rules on what the notice must say were proposed and then withdrawn.

Jurisdiction: US (Illinois)

Status: In force

Key date: Signed 2024-08-09 (Public Act 103-0804); effective 2026-01-01; notice rules proposed 2026-05-15, withdrawn 2026-06-02

Issued by: Illinois General Assembly / Department of Human Rights

Instrument: State statute (amends the Human Rights Act)

Scope: Illinois employers of any size using AI in recruitment, hiring, promotion, discipline or discharge

Key obligations: No use of AI that has a discriminatory effect on a protected class, including using zip code as a stand-in; notice to employees and applicants when AI is used in a covered decision

If you use AI in your business: Applies to you if you have even one employee in Illinois and use AI anywhere in hiring or managing staff. Tell applicants and employees that you use it. What the notice must contain is unsettled, because the state withdrew its draft rules in June 2026. Discriminatory effect is the violation whether or not you intended it, and people can bring claims through the Human Rights Act.

Related issues: AI bias in hiring & employment

Sources: CASRAI: Illinois HB 3773, an AI notice duty with no rules behind it; National Law Review: Illinois anti-discrimination law to address AI goes into effect on 1 January 2026

China — Deep Synthesis Provisions

Requires labeling of AI-generated content, real-name authentication for deep-synthesis users, and content-moderation duties for providers.

Jurisdiction: China

Status: In force

Key date: Effective 2023-01-10

Issued by: CAC / MIIT / MPS

Instrument: Administrative provisions

Scope: Deep-synthesis service providers and users in China

Key obligations: Label synthetic content, verify user identity, conduct risk assessment for deepfakes

If you use AI in your business: Both providers and users (deployers) of deep-synthesis tools must label AI-generated content; users must authenticate with real names before generating synthetic media.

Related issues: Deepfakes & election misinformation; AI 'nudification' & non-consensual intimate imagery; AI surveillance & workplace monitoring

Sources: Library of Congress — China Deep Synthesis Provisions

China — Algorithmic Recommendation Provisions

Regulates recommendation systems: transparency, user controls, protection of minors and workers, and limits on manipulation and price discrimination.

Jurisdiction: China

Status: In force

Key date: Effective 2022-03-01

Issued by: CAC

Instrument: Administrative provisions

Scope: Algorithmic-recommendation service providers in China

Key obligations: Algorithm transparency, opt-out, worker protections, minor protections, anti-manipulation

If you use AI in your business: Binds recommendation-service providers; deployer-side duties include offering users an opt-out, protecting workers from algorithmic management harms, and minor-protection defaults.

Related issues: AI surveillance & workplace monitoring; AI social scoring & manipulation

Sources: Library of Congress — China Deep Synthesis (comparative CAC framework)

China — Rules for human-like AI companions

The Interim Measures for the Administration of AI Anthropomorphic Interactive Services cover AI that keeps up an emotional relationship by imitating a person's personality and way of talking. Customer service bots, work assistants and tools for education and research are outside it.

Jurisdiction: China

Status: In force

Key date: Issued 2026-04-10; effective 2026-07-15

Issued by: Cyberspace Administration of China and four other departments

Instrument: Administrative measures (interim)

Scope: Providers of human-like emotional interaction services offered to the public in China

Key obligations: Guardian consent for users under 14; no virtual intimate-relationship services for minors; a reminder when a session passes two hours; stepped interventions when a user shows signs of harmful use

If you use AI in your business: Applies if you offer a companion-style AI product to the public in China, whether you built the model or not. An ordinary customer service chatbot is outside it.

Related issues: AI social scoring & manipulation

Sources: Bird & Bird: China's new regulations on AI anthropomorphic interactive services; State Council of China: government policy moves from the past week (15 April 2026)

UK — Online Safety Act & pro-innovation framework

Sector-regulator approach (no single AI law); Online Safety Act addresses harmful AI-generated content on platforms; ICO/CMA apply existing law to AI.

Jurisdiction: UK

Status: In force (framework); AI-specific evolving

Key date: OSA 2023-10; AI White Paper 2023 (pro-innovation); no AI bill in the 2026 King's Speech

Issued by: DSIT / ICO / CMA / Ofcom

Instrument: Statute + principles-based guidance

Scope: Platforms, online services, and AI deployers in the UK

Key obligations: Duty of care for harmful content; data-protection application to AI; consumer-protection enforcement

If you use AI in your business: Sector-regulator approach with no single AI law. Platforms owe a duty of care for harmful content; existing UK GDPR and consumer law apply directly to SMEs deploying AI — so your duties come from data-protection law, not an AI statute.

Related issues: Deepfakes & election misinformation; AI surveillance & workplace monitoring; AI market concentration & compute; Agentic AI & autonomous-action liability

Sources: European Commission — AI Act framework (comparative); IAPP: King's Speech signals diffuse UK digital policy agenda, but no AI bill

Japan — Soft-law AI framework (non-binding)

Deliberately avoids risk-tiered binding rules. Three non-binding pillars: the AI Promotion Act (strategy), the AI Guidelines for Business (voluntary governance), and agency-specific guidance.

Jurisdiction: Japan

Status: In force (non-binding)

Key date: AI Promotion Act 2025; AI Guidelines for Business v1.2, March 2026; AI Basic Plan revised 2026-07-14

Issued by: Digital Agency / MIC / ministries

Instrument: Non-binding statute + voluntary guidelines

Scope: Businesses developing or using AI in Japan (voluntary)

Key obligations: Voluntary risk management, transparency, human oversight, stakeholder participation — no penalties

If you use AI in your business: Non-binding. The AI Promotion Act sets strategy; the AI Guidelines for Business ask deployers to voluntarily manage AI risks. No obligations, no penalties — a deliberate philosophical contrast to the EU's risk-tiering regime.

Related issues: AI & scientific research integrity; AI market concentration & compute; Agentic AI & autonomous-action liability

Sources: IBA — Japan's emerging framework for responsible AI; MIC Japan — AI Guidelines for Business v1.2 (PDF); White & Case: Japan's first AI legislation becomes law

South Korea — AI Basic Act

Comprehensive law defining high-risk and generative AI, with transparency, safety, and operator-responsibility duties. First major Asian AI law to take effect.

Jurisdiction: South Korea

Status: In force

Key date: Passed 2024-12-26; in force 2026-01-22; fines held back for at least a year

Issued by: National Assembly / MSIT

Instrument: Statute (framework)

Scope: High-risk and generative AI providers/operators in Korea

Key obligations: Transparency, safety, impact assessments, operator responsibilities

If you use AI in your business: In force since Jan 2026. 'Operators' — including businesses that deploy high-risk or generative AI — must ensure transparency, safety, and run impact assessments. Deployers are explicitly in scope, not just developers. The duties apply now, but the ministry has said it will hold back administrative fines for at least a year from January 2026.

Related issues: AI bias in hiring & employment; AI-assisted cheating in education; AI & scientific research integrity

Sources: MSIT Korea — AI Basic Act press release; eWeek: foreign AI firms face 2026 compliance window under South Korea AI law

Brazil — AI Bill (PL 2338/2023)

Risk-based framework modelled on the EU AI Act, classifying systems minimal-to-excessive. Passed the Senate unanimously; pending in the Chamber of Deputies, where a special committee has yet to report. A government bill on a national AI governance system (PL 6237/2025) was attached to it in March 2026.

Jurisdiction: Brazil

Status: Proposed

Key date: Senate approved 2024-12-10; in the Chamber of Deputies since 2025, awaiting a committee opinion (no vote as of 2026-10)

Issued by: Federal Senate / Chamber of Deputies

Instrument: Bill (not yet law)

Scope: AI developers and deployers in Brazil

Key obligations: Risk classification, impact assessments, rights for affected persons, prohibited practices

If you use AI in your business: Not yet law. Would bind both developers and deployers of AI in Brazil under a risk-tiered framework; deployers of high-risk systems would face impact-assessment and disclosure duties.

Related issues: Deepfakes & election misinformation; AI bias in hiring & employment; AI social scoring & manipulation

Sources: Deep Lex — Brazil AI regulation tracker; artificialintelligenceact.com — Brazil AI Act; CASRAI: Brazil's AI bill PL 2338, approved, not law

US Copyright Office — AI guidance

Holds that purely AI-generated works lack human authorship and cannot be copyrighted; works with sufficient human authorship may be protectable.

Jurisdiction: US (federal)

Status: In force

Key date: Guidance 2023-03; ongoing review

Issued by: US Copyright Office

Instrument: Administrative guidance / registration policy

Scope: AI-assisted creative works in the US

Key obligations: Disclose AI-generated content in registration; human authorship required

If you use AI in your business: Not a provider/deployer rule. Governs whether AI-assisted works can be registered for copyright; binds creators and the Copyright Office, not AI deployers as such.

Related issues: Generative AI & copyright infringement; AI voice cloning of artists

Sources: Harvard Law — AI, copyright & the Drake case

US FDA — AI/ML Software as a Medical Device (SaMD)

Mature, evolving framework governing AI-enabled medical devices: Predetermined Change Control Plans let manufacturers pre-authorize model updates, plus lifecycle transparency and monitoring guidance.

Jurisdiction: US (federal)

Status: In force (evolving)

Key date: PCCP final guidance 2024-12-03; lifecycle draft 2025-01-07

Issued by: FDA (Center for Devices and Radiological Health)

Instrument: Guidance (non-binding but expected) + existing device authorities

Scope: Manufacturers of AI-enabled device software functions in the US

Key obligations: Predetermined Change Control Plan, lifecycle transparency, performance monitoring, labelling

If you use AI in your business: Binds medical-device manufacturers, not hospitals or clinics that deploy the devices. Manufacturers must file a PCCP and follow lifecycle transparency/monitoring guidance. Deployers (clinicians) are regulated through existing medical-device and practice rules, not the PCCP.

Related issues: AI & scientific research integrity

Sources: FDA — Artificial Intelligence in Software as a Medical Device; AI SaMD Playbook — PCCP & lifecycle guidance

More from Areactor

Longer essays behind some of the ideas this dataset only summarizes.