AccuroAI
Book a demo
← Blog·Pillar Hub15 min read

AI Compliance by Market: Which Regime Binds You

Which law binds your use of AI depends on the market, and in most markets it is a data protection statute, not an AI statute. This hub maps the binding regime, the next dated change and our guide for eleven market groups, from the EU and the UK to India, the Middle East, Africa and Brazil. It then shows how one 20-control library produces evidence for NIST AI RMF, ISO/IEC 42001, the EU AI Act and the GDPR.

A
AccuroAI Editorial
Editorial
May 30, 2026
On this page

AI compliance starts with one question: which law binds your use of AI in each market where you have staff, customers or data? In most of the world the answer is a data protection statute, not an AI statute. Of the eleven market groups mapped below, our guides found AI-specific law in force in only three: the EU's AI Act, Vietnam's Law on Artificial Intelligence, and a handful of US state laws. Everywhere else, privacy law already reaches prompts, uploads and agent actions, and it asks for nearly the same four things: an inventory of AI tools, a contract with each provider, a record of what personal data went where, and a route for people to challenge automated decisions. This hub maps the binding regime for each group, links the guide that covers it, and then shows how one control library produces evidence for all of them.

Last verified: 13 November 2026, against EUR-Lex (Regulation (EU) 2026/1744, the GDPR, Council Decision (EU) 2026/1080 and the convention text in the Official Journal), EFTA's EEA-Lex, Canada's LEGISinfo and OSFI, the Texas and Colorado legislature pages, Brazil's Câmara dos Deputados, the Turkish regulator's translation of Law No. 6698, NIST and OWASP. Every other cell in the regime table is carried from the linked guide as that guide last verified it; the primary pages were not re-opened today. The Council of Europe's treaty pages refused our requests, so the convention section states only what EUR-Lex shows.

Which AI and privacy regime binds you in each market?

Read the table by row, one row for each place where you employ people, serve customers or process their data. The second column names the instruments exactly, because the names are where most summaries go wrong. The third gives the next change with a date attached. Where a cell says "none dated", we found a bill or a consultation but no commencement date.

Market group What binds enterprise AI use today Next dated change What it asks if you use ChatGPT, Copilot or agents Our guide
EU and EEA GDPR, Regulation (EU) 2016/679. AI Act, Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744. Italy adds Law 132/2025 and Legislative Decree 160/2026. Norway: the GDPR applies through the EEA Agreement, but the AI Act is not yet incorporated into it. Switzerland sits outside both and has its own Federal Act on Data Protection (FADP) 2 December 2026: machine-readable marking under Article 50(2) for generative systems already on the market before 2 August 2026, and two new Article 5 prohibitions. 2 December 2027: Annex III high-risk duties An Article 28 contract and an Article 30 record entry for every AI tool, a transfer mechanism, a DPIA before a rollout at scale, and a chatbot that says it is an AI GDPR and enterprise AI; Copilot DPIA findings; AI Act enforcement tracker; Italy's AI law
United Kingdom UK GDPR and the Data Protection Act 2018, both amended by the Data (Use and Access) Act 2025. Articles 22A to 22D on automated decisions have been in force since 5 February 2026. There is no UK AI Act None dated in statute. The regulator lists final guidance on automated decision-making for winter 2026 Four working safeguards for any significant decision taken with no meaningful human involvement: information, representations, human intervention and a way to contest UK AI governance after the 2025 Act
Canada PIPEDA (S.C. 2000, c. 5). In Quebec, section 12.1 of the private sector Act, added by Law 25 and in force since 22 September 2023. No AI statute: AIDA was never passed, and Bill C-36 is a privacy bill that has completed first reading only 1 May 2027: OSFI Guideline E-23 on model risk takes effect for federally regulated financial institutions You stay accountable for personal information sent to an AI vendor. In Quebec, tell the person when a decision is exclusively automated and offer review by a staff member Canada after AIDA
United States A patchwork of state laws beside federal sector rules such as HIPAA and SR 11-7. In force since 1 January 2026: Texas TRAIGA (HB 149), Illinois HB 3773 on AI in employment, and California SB 53 and AB 2013 for model developers. Proposals for federal preemption have not settled the matter 1 January 2027: Colorado SB 26-189, California's CCPA regulations on automated decision-making technology, and New York's RAISE Act. 1 July 2027: California SB 947 Notice when AI is used in employment decisions in Illinois. From 2027 in Colorado: notice, an explanation within 30 days of an adverse outcome, and meaningful human review US state AI law tracker; Texas TRAIGA safe harbor
Australia and New Zealand Australia: the Privacy Act and the Australian Privacy Principles, which the regulator says apply to all uses of AI involving personal information. New Zealand: the Privacy Act 2020, including information privacy principle 3A on indirect collection, in force since 1 May 2026 10 December 2026: APP 1.7 to 1.9, inserted by the Privacy and Other Legislation Amendment Act 2024, start to apply From that date, a privacy policy that names the kinds of decisions computer programs make or substantially shape, and the kinds of personal information they use Australia's automated decision disclosure rule
Singapore Personal Data Protection Act 2012. The Model AI Governance Framework for Agentic AI (version 1.5, 20 May 2026) is voluntary, and the PDPC's advisory guidelines on AI are not legally binding 7 October 2027: the MAS Guidelines on AI Risk Management take effect for financial institutions Reasonable security arrangements for personal data an agent touches, and notice to the PDPC within three calendar days of assessing a breach as notifiable Singapore's agentic AI framework, mapped to controls
Southeast Asia Malaysia: Personal Data Protection Act 2010 (Act 709), amended by Act A1727. Indonesia: Law No. 27 of 2022. Vietnam: Law 91/2025/QH15 with Decree 356/2025/NĐ-CP, and the AI Law 134/2025/QH15 with Decree 142/2026/NĐ-CP. The Philippines: Data Privacy Act of 2012 and its Implementing Rules. Thailand: Personal Data Protection Act B.E. 2562 (2019), which our guide read in an unofficial translation only 1 March 2027: by our calculation from the law's transition article, the deadline for AI systems already operating in Vietnam before 1 March 2026 (1 September 2027 in health, education and finance) One breach matrix, because the clocks start at different moments. In Vietnam, classify each system and report serious AI incidents. In the Philippines, register systems that make automated decisions Southeast Asia: six markets, one table
India Digital Personal Data Protection Act, 2023 and the Digital Personal Data Protection Rules, 2025, which commence in tranches. The duties for ordinary Data Fiduciaries are not yet in force, and sectoral localization rules continue to apply 13 May 2027: duties for every Data Fiduciary (some advisers count 14 May). The tranche dated 13 November 2026 concerns Consent Manager registration only From May 2027: a valid contract with every processor, AI providers included, one year of access logs, and complete, accurate and consistent data behind decisions that affect a person India's DPDP Act and AI
Middle East UAE: Federal Decree by Law No. (45) of 2021, with separate regimes in the DIFC and ADGM. Saudi Arabia: Personal Data Protection Law, Royal Decree No. (M/19), with its Implementing Regulation and transfer regulation. Qatar: Law No. (13) of 2016. Egypt: Law No. 151 of 2020. Türkiye: Law No. 6698 (KVKK). Our guide found no AI Act in force in any of them None dated that we verified. Law firms put the end of Egypt's one-year regularization period at 31 October or 1 November 2026; confirm it with local counsel Treat a prompt sent to a tool hosted abroad as a transfer: a license or permit in Egypt, a written risk assessment in defined cases in Saudi Arabia, and a standard contract notified within five business days in Türkiye Middle East transfer rules compared
Africa Nigeria: Nigeria Data Protection Act, 2023 and the GAID 2025 directive. Kenya: Data Protection Act, No. 24 of 2019. South Africa: Protection of Personal Information Act, No. 4 of 2013 (POPIA). None of the three has an AI statute 31 March each year: Nigeria's audit return for the two higher registration tiers. Kenya's Artificial Intelligence Bill, 2026 was at second reading stage in October, with no date A right over solely automated decisions in all three, an impact assessment (filed with the regulator in 15 listed cases in Nigeria, submitted 60 days ahead in Kenya), and breach notice within 72 hours in Nigeria and Kenya AI rules in Nigeria, Kenya and South Africa
Brazil LGPD (Lei 13.709/2018), with ANPD Resolution 15/2024 on incidents and Resolution 19/2024 on international transfers. PL 2338/2023 is a bill, not law None dated. The bill is still awaiting its rapporteur's report in a special committee of the Câmara dos Deputados A transfer mechanism for each AI vendor abroad, review on request of solely automated decisions under Article 20, and notice of a relevant incident within three business days Brazil: LGPD, the ANPD and the AI bill

Three patterns repeat. Automated decisions are the common thread: almost every row carries a right or duty about them, and the texts differ on whether advisory tools count. Transfers come second, since a prompt sent to a model hosted abroad is an export in most of these regimes. Breach clocks are third.

Two naming points for the first row. Norway applies the GDPR as national law through the EEA Agreement, yet EFTA's EEA-Lex database still shows the AI Act as "under scrutiny for incorporation into the EEA Agreement", so it does not bind there yet. Switzerland is in neither the EU nor the EEA; its statute is the FADP, in force since 1 September 2023, and its regulator has said that law applies directly to AI-supported processing.

Where does the Council of Europe AI convention stand?

The Council of Europe Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law (CETS No. 225) is the one international treaty in this area, and we can report only part of its status today. On 13 November 2026 the Council of Europe's treaty office pages refused our requests, so we did not read the chart of signatures and ratifications. We therefore make no statement on whether the convention is in force or on which states have signed or ratified it.

EUR-Lex shows the following. The convention was done at Vilnius on 5 September 2024 and signed for the European Union that day. Council Decision (EU) 2026/1080 of 21 April 2026 approved it on the Union's behalf, and Article 3 of that decision says it will be implemented in the Union exclusively through the AI Act, with other Union law where applicable. Under Article 30(3) of the convention, entry into force needs five signatories, at least three of them Council of Europe member states, to have consented to be bound, followed by a three-month waiting period.

For a company in the EU, then, the convention adds no duty beyond the AI Act. Elsewhere, check the treaty office chart before repeating any claim that a particular country has ratified.

Which frameworks turn those laws into one program?

Laws say what you must be able to show. Frameworks give you a structure for showing it, and five of them anchor most enterprise programs.

Framework Status What it covers
NIST AI RMF (version 1.0, 26 January 2023; Generative AI Profile, NIST AI 600-1, 26 July 2024) Voluntary US framework Four functions: Govern, Map, Measure, Manage
ISO/IEC 42001 (2023) Certifiable management system standard An AI management system, with Annex A controls A.2 to A.10
EU AI Act (Regulation (EU) 2024/1689) Binding EU law, applying in phases from 2025 to 2028 Risk tiers (prohibited, high-risk, transparency duties, minimal) with separate obligations for providers and deployers
GDPR (Regulation (EU) 2016/679) Binding EU law, and the model for many of the privacy statutes in the regime table Contracts, records, DPIAs, breach notification, transfers and automated decisions
OWASP Top 10 for Agentic Applications for 2026 Industry list, published 9 December 2025 Ten risks specific to agentic systems

Sector frameworks sit on top: SOC 2, HIPAA, PCI DSS and SR 11-7 for financial services. The practical point is that a single control library can feed all of them, which is the subject of our unified crosswalk.

One correction to a claim this page used to make. Article 40 of the AI Act gives a presumption of conformity only to harmonized standards cited in the Official Journal, and when our enforcement tracker last checked, none had been cited. ISO/IEC 42001 certification is useful evidence. It does not yet carry that presumption.

The 20-control unified library

The crosswalk reduces the frameworks to twenty operational controls:

  1. Written AI Acceptable Use Policy
  2. Named AI Governance Committee
  3. AI literacy program
  4. AI inventory (continuous)
  5. AI Bill of Materials (AIBOM)
  6. Risk classification per system
  7. Inline prompt inspection
  8. Inline response inspection
  9. Tool allowlisting
  10. Capability-scoped tokens
  11. Provenance logging (every interaction)
  12. A2A signing and message inspection
  13. Kill switches (per system, per tool)
  14. AI incident response runbook
  15. Adversarial testing (red-team)
  16. Bias and fairness evaluation
  17. Data governance (training and operational)
  18. Transparency and disclosure to users
  19. Human oversight design
  20. Continuous monitoring and post-market surveillance

Set the list beside the regime table and the overlap is plain. The inventory (4) is the first step in every market guide we have written. Provenance logging (11) is what a breach assessment, an Article 30 record and a Colorado explanation all draw on. Transparency (18) and human oversight (19) answer the automated-decision rules from Quebec to Kenya.

The EU AI Act timeline after the Digital Omnibus on AI

Regulation (EU) 2026/1744 of 8 July 2026, published in the Official Journal on 24 July 2026, moved the high-risk dates and left the rest in place. The operative timeline:

  • 2 February 2025: Article 5 prohibitions apply. So does Article 4, which the Omnibus reworded: providers and deployers must "take measures to support the development of AI literacy" of their staff, with no duty to guarantee a specific level.
  • 2 August 2025: obligations for general-purpose AI model providers apply.
  • 2 August 2026: the Act applies in general. The Commission can fine general-purpose model providers, and Article 50 transparency applies.
  • 2 December 2026: the Article 50(2) marking deadline for generative systems placed on the market before 2 August 2026, and the two new Article 5 prohibitions.
  • 2 August 2027: general-purpose models placed on the market before 2 August 2025 must comply.
  • 2 December 2027: Annex III high-risk obligations, moved from 2 August 2026.
  • 2 August 2028: Annex I high-risk obligations, moved from 2 August 2027.

The Omnibus on AI does not amend the GDPR. See the dates-only reference and what has applied since 2 August 2026.

All AccuroAI posts on AI compliance

The fifteen market guides are linked in the regime table above. The rest of the library:

EU AI Act

Frameworks and evidence

Governance and buying

What to do this quarter

  1. Mark your rows in the regime table, one for each market where you have employees, customers or data subjects, and read the guide for each.
  2. Diary the two December dates if they reach you: 2 December 2026 for Article 50(2) marking in the EU, and 10 December 2026 for the Australian privacy policy disclosure. Colorado and California follow on 1 January 2027.
  3. Pick the anchoring frameworks. We suggest ISO/IEC 42001 and NIST AI RMF as the baseline, with the AI Act and the GDPR layered on for in-scope systems and data.
  4. Build the 20-control library from the crosswalk, then inventory existing evidence control by control and list the gaps.
  5. Brief internal audit on the single evidence pack, so one set of records answers several frameworks.

Two free tools shorten the first month. Our AI governance maturity assessment takes about twelve minutes and returns a ranked list of gaps, which tells you where to start on the library. The AI acceptable use policy template gives you the written rule nearly every regime above assumes, and its Appendix D now carries jurisdiction rows for the UK, Australia, Quebec, Italy, Singapore, Vietnam, Brazil, Nigeria and Saudi Arabia.

Which parts of this can software carry, and which stay with people?

No product makes an organization compliant with any law or framework on this page. Choosing a lawful basis, writing a DPIA, signing a transfer contract and deciding whether a human review is meaningful are legal and management judgments.

Software can carry the facts those judgments rest on. AccuroAI recognizes 1,400+ AI tools in the browser and on managed devices, with the signed-in account and plan, which is the inventory every row of the regime table starts from. It redacts or blocks sensitive data in line, using 40+ classifiers, before a prompt reaches a model, so a tool with no contract or transfer basis does not receive personal data. Each event is logged, and the evidence is mapped to 8 frameworks.

FAQ

Which markets have an AI-specific law in force today?

Among the groups our guides cover: the EU, through the AI Act; Vietnam, through Law 134/2025/QH15, in force since 1 March 2026; and several US states, including Texas and Illinois. Our guides found no AI statute in force in the UK, Canada, Brazil, Nigeria, Kenya, South Africa or the seven Middle East regimes they compared. Singapore's agentic framework is voluntary.

Does the EU AI Act apply in Norway, Switzerland or the UK?

Not in Norway yet, because the Act has not been incorporated into the EEA Agreement. It is not UK law either, though a UK company can fall under it when it places AI systems on the EU market or their outputs are used there. Switzerland regulates AI-supported processing of personal data through the FADP.

Has the EU AI Act been delayed?

In part. Annex III high-risk obligations moved to 2 December 2027 and Annex I to 2 August 2028. Enforcement against general-purpose model providers and Article 50 transparency kept their 2 August 2026 date. The older prohibitions kept theirs, two new ones apply from 2 December 2026, and the AI literacy duty was reworded.

Is ISO 42001 certification required?

No. Certification is optional, though procurement teams increasingly ask for it. It does not yet give a presumption of conformity with the AI Act, for the reason set out above.

What is the minimum baseline for an enterprise starting now?

The 20-control library. A first pass typically lands somewhere between a quarter and a half of the controls, and a reasonable target is 18 to 20 within 12 months. Start with the inventory.

This page is not legal advice. It summarizes our own guides, each of which names its sources and the date it was checked; ask counsel in each market how the rules apply to you.

Sources

More from the blog
See AccuroAI in action.

Book a 30-minute demo and see how security teams use AccuroAI to discover, govern, and protect every AI asset across their organization.

Book a demoStart a free pilot

15 enterprises secured · under 38ms p99 · Deployed in under 30 minutes · 72-hour pilot