AI in the Americas and Europe, Country by Country

Key takeaways

• People adopt AI faster than firms do, almost everywhere. Microsoft's June 2026 telemetry puts working-age generative-AI use at 49.9% in Ireland, 49.6% in France, 43.7% in the UK, 38.5% in Canada and 33.0% in the US (21st in Microsoft's table). Official surveys put firm adoption at 17-20% in the US, 19.2% in Canada and 19.95% across the EU. The gap between those two numbers is where most businesses still have processes to automate. Denmark and Finland are the exceptions: there, more firms use AI than individuals do.
• The EU has moved its AI Act dates but kept the rules. The AI Omnibus took effect on 27 July 2026. High-risk uses such as hiring and credit scoring (Annex III) now apply from 2 December 2027, and AI built into regulated products (Annex I) from 2 August 2028. Transparency duties (AI disclosure, deepfake labels, machine-readable marking) and AI Office enforcement over general-purpose models began on 2 August 2026.
• Washington wants to preempt state AI laws, but state dates still govern US businesses. EO 14365 (December 2025) ordered a federal task force to litigate against state AI laws. No court ruling on it was found. Texas TRAIGA has applied since 1 January 2026, according to earlier sources (the enrolled text gives no date). California's automated-decision rules, Colorado's rewritten law (SB26-189) and Colorado's Chatbot Safety Act all start on 1 January 2027. A federal court order (X.AI v. Weiser, 27 April 2026; the US joined xAI as plaintiff) bars Colorado's Attorney General from starting enforcement until 14 days after a preliminary-injunction ruling; whether it reaches SB26-189 is arguable.
• US compute spending is in a class of its own. Alphabet, Meta and Amazon spent $228.1 billion on capex in H1 2026 (+81%), and Microsoft spent $116.0 billion in its year to June 2026. Europe's main answer is an AI gigafactories call. It closes on 12 November 2026 and expects more than EUR 20 billion of private investment, alongside a reported EUR 10 billion of public money.
• Latin America is still at the strategy stage. Brazil's AI bill, passed by the Senate in December 2024, has been "awaiting the rapporteur's opinion" in a Chamber special committee since March 2025; the rapporteur expected a vote only after the October 2026 elections (MobileTime, 24 Aug 2026, not re-read on 12 October; the Chamber record sets no vote date). Mexico's National AI Plan 2026 is reported to centre on a 314-petaflop public supercomputer. Chile's bill is in the Senate, and Colombia runs the CONPES 4144 policy. For Argentina, the research found only a stalled OpenAI letter of intent and a usage figure.
• The UK has reorganised under a new Prime Minister. Andy Burnham has been Prime Minister since 20 July 2026, when Kanishka Narayan was also appointed Minister for AI, attending Cabinet. GOV.UK says DSIT is being replaced by DBIST, DCMS and the Cabinet Office. A UK Parliament statement (May 2026; not re-read on 12 October) puts private investment committed to five AI Growth Zones at more than GBP 28 billion; GOV.UK separately cites GBP 24.25 billion across all sectors in the month to 20 November 2025. The UK GDPR automated-decision articles (22A-22D) have applied since 5 February 2026.
• Existing laws are doing the enforcing. The 2026 Grok deepfake case drew action under the EU's Digital Services Act, Irish data-protection law and Canadian privacy law, plus a UK Ofcom investigation and US state attorney-general action. None of these is an AI statute.
• Data are thin below the largest economies. For Poland, Estonia, Spain, the Nordics, the Netherlands, Ireland, Switzerland, Chile, Colombia and Argentina, the research has little beyond adoption statistics or a single policy document. Those gaps are stated, not filled.

All facts in this document are as of 12 October 2026; key legal, investment and statistical facts were re-checked against primary sources on 12 October 2026, and facts not re-read keep their 11 October reading. Where the text says a page was checked or read on a date, the value was confirmed on the live page then. This is research, not legal advice. Labels: secondary means the fact comes from news, law-firm or encyclopaedia coverage rather than the primary text; single-source means only one source supports it; headline only means only a news headline was seen. Related pages: country overview, Regulation and security, India, Asia-Pacific, Middle East and Africa.

Regional summary table

How to read the table. The Microsoft column is the share of people aged 15 to 64 who used a generative-AI product in June 2026, estimated from anonymised telemetry. "n/a" means the country is not in the report's country table. The values for Denmark, Finland, Poland, Chile, Colombia and Argentina, and the US rank, were read from the report on 11 October 2026 (Microsoft AI Diffusion Report Q2 2026).

The enterprise column combines four official surveys that ask different questions:

• Eurostat counts enterprises with ten or more staff that use at least one AI technology. Its headline covers enterprises in NACE divisions 10-95, excluding finance (K): agriculture and mining are outside the scope, utilities are inside (Eurostat API, isoc_eb_ai, 2025).
• Statistics Canada counts businesses that used AI to produce goods or deliver services in the past year.
• The US Census widened its question in November 2025 to cover AI use in any business function.
• The UK's ONS Business Insights survey (BICS) asks businesses whether they use some form of AI technology (ONS, 8 Jan 2026).

The four are indicative, not strictly comparable. The IMF AI Preparedness Index runs from 0 to 1 and uses 2023 data, the only vintage its API returns. The values for Denmark, Finland, Norway, Poland, Estonia, Belgium, Luxembourg, Chile, Colombia and Argentina were read from the API on 11 October 2026 (IMF DataMapper, AI_PI). TOP500 counts refer to the world's top 100 supercomputers in June 2026 (TOP500, June 2026).

Three patterns stand out. Individual use and firm adoption diverge: France and Ireland lead Europe on Microsoft's measure, but only 18-20% of their firms use AI, while Denmark leads on firms at 42% with individual use of 32.4%. The US does not lead on adoption by either measure. Eleven European economies (including Austria and Hungary) and Canada sit above it on Microsoft's list, where it ranks 21st. Its lead is in capital, compute and frontier models. Regulation splits three ways: the EU's single binding law, the US contest between federal and state power, and a group (the UK, Canada and most of Latin America) that relies on strategies, data-protection law and sector regulators.

United States: preemption pressure, a state patchwork and the gigawatt economy

Federal: deregulation at home, pressure on the states

Federal policy since January 2025 has removed federal limits on AI and challenged state ones. EO 14179 (23 January 2025) revoked the Biden-era EO 14110, and "America's AI Action Plan" followed on 23 July 2025 with more than 90 actions (secondary) (Wikipedia: 2025 in artificial intelligence).

Four orders signed since November 2025 define the current framework (Federal Register):

• EO 14363, "Launching the Genesis Mission" (signed 24 November 2025) directs the Department of Energy to build an AI science platform on federal datasets. No budget was found (Federal Register: EO 14363; MoFo).
• EO 14365, "Ensuring a National Policy Framework for Artificial Intelligence" (signed 11 December 2025, published 16 December) (Federal Register: EO 14365). One secondary source numbered it 14257. The Federal Register number is 14365.
• EO 14409, "Promoting Advanced Artificial Intelligence Innovation and Security" (signed 2 June 2026; 91 FR 34565): frontier-model cyber defence via CISA, Treasury and the NSA; a classified cyber-capability benchmark; a voluntary framework under which developers give the government up to 30 days of pre-release access. Sec. 3(c) rules out mandatory licensing or preclearance. No implementation outputs were found (Federal Register: EO 14409).
• EO 14434, "Inaugurating the Era of Super Intelligence" (signed 29 September 2026; 91 FR 63129): agencies write "Super Intelligence" (SI) for "AI" in official communications as far as the law allows, and the President's science and technology assistant (APST) proposes legislative language for a federal SI definition within 60 days (about 28 November 2026) (Federal Register: EO 14434).

EO 14365 is the order US businesses need to understand. Its text (re-read 11 Oct 2026) directs the Attorney General to set up, within 30 days, an AI Litigation Task Force to challenge state AI laws, including on interstate-commerce and preemption grounds. It gives Commerce 90 days (to about 11 March 2026) to publish an evaluation naming "onerous" state laws, and makes states with such laws ineligible for BEAD broadband non-deployment funds to the extent federal law allows (Federal Register: EO 14365; Paul Hastings; Introl). The task force was formally established on 9 January 2026 (secondary) (Weintraub). The order also calls for a legislative recommendation for a federal framework that preempts conflicting state laws, but says it should not propose preempting state laws on child safety, AI compute and data-centre infrastructure, or state procurement and use of AI (Federal Register: EO 14365). The research found no court ruling on the order and no published Commerce list. The order works through litigation, funding conditions and proposed legislation; it does not itself repeal state statutes, so state laws remain enforceable unless a court rules otherwise.

The personnel shifted during 2026; see People for profiles. David Sacks ended his role as White House AI and crypto czar in March and now co-chairs the President's Council of Advisors on Science and Technology with Michael Kratsios, director of the Office of Science and Technology Policy. Sacks says the council's agenda includes a national framework to replace state rules (Daily Signal, 26 Mar 2026). Sriram Krishnan left the White House AI adviser post on 30 June 2026 and later said "there will not be an FDA for AI" (secondary, Wikipedia citing the FT) (Wikipedia: Sriram Krishnan). Congress has passed no preemption law. The most concrete text found is Senator Mark Warner's discussion draft of an AI AGENT Act (29 June 2026), which covers consumer-facing agents (secondary) (Wikipedia: Regulation of artificial intelligence). An Obernolte–Trahan bipartisan discussion draft (June 2026), described as a three-year freeze on state rules on how AI is built, reportedly stalled (single-source secondary) (CASRAI, Sep 2026). The standards layer is voluntary. NIST's AI Risk Management Framework (January 2023) and Generative AI Profile (26 July 2024) are still the references (NIST), and NIST's Center for AI Standards and Innovation (CAISI), which NIST's site now calls CAISSI (Center for Advancing Innovation and Standards for Super Intelligence), runs an "AI Agent Standards Initiative" (NIST CAISSI); the rename date is not established. The NIST framework also carries legal weight in Texas (see the state table below).

Federal power over models: export controls, vetting and procurement

Access to a model is now a regulatory variable. On 12 June 2026 a US export-control directive suspended all access by foreign nationals to Anthropic's Claude Fable 5 and Mythos 5. Anthropic disabled both models for every customer until the controls were lifted on 30 June, and access returned on 1 July (Anthropic, 12 Jun 2026; Al Jazeera, 1 Jul 2026). OpenAI said the government would vet users of GPT-5.6 Sol (headline only) (Washington Post).

Procurement decisions also shape which vendors federal contractors can use. The Pentagon designated Anthropic a supply-chain risk in February 2026. A judge blocked the designation in March, and an appeals court reportedly upheld the designation on 25 September (both headline only). On 5 October the BBC reported that the Department of Defense "has ceased the use of Anthropic products" (BBC). Meanwhile, xAI, OpenAI, Anthropic and Google each held Defense contracts with $200 million ceilings from 2025 (secondary) (CBS News). On the civilian side, Anthropic committed $150 million over three years to the Genesis Mission, covering more than 15 agencies including NASA, NIH and NSF (Anthropic, 8 Oct 2026). Civilian buying runs through GSA's OneGov deals: xAI offers Grok 4 and Grok 4 Fast at $0.42 per organisation (agency) for 18 months, valid until March 2027 (GSA, 25 Sep 2025), and a new 27-month OpenAI phase from about 1 October 2026 gives 50% off consumption-based usage. GSA says OneGov has saved about $1.68 billion, $1.4 billion of it from AI agreements, across about 3.5 million federal employees (GSA, 10 Sep 2026). No federal AI use-case counts were retrievable.

Copyright is being settled case by case. In NYT v. Microsoft and OpenAI, the Justice Department filed a brief in September 2026 rejecting the view that training on copyrighted text is infringement in itself (secondary) (Wikipedia: NYT v. Microsoft and OpenAI). Anthropic's US$1.5 billion settlement with authors received final approval on 20 July 2026 (secondary) (Bloomberg Law, settlement).

States: four models of AI law

Sources: Texas (Texas Legislature, HB 149 enrolled text, primary; Nelson Mullins; Wikipedia: TRAIGA); California SB 53 (Clifford Chance; NatLawReview, SB 53 obligations now and in 2027, both checked 11 Oct 2026); California ADMT (CPPA, approved regulation text, primary); Colorado (Colorado General Assembly, SB26-189, checked 11 Oct 2026; Colorado Attorney General, ADMT and Chatbot Safety rulemaking, checked 11 Oct 2026; Mondaq/DGL; CourtListener: X.AI LLC v. Weiser); New York (Wikipedia: RAISE Act; SecurePrivacy: New York AI law guide, secondary).

Colorado shows where state law is heading. Its 2024 AI Act imposed a duty of care against algorithmic discrimination, backed by impact assessments, and was first delayed to 30 June 2026 (Greenberg Traurig). In May 2026 the legislature repealed it and replaced it with a disclosure-and-recourse model that drops the duty of care and the impact assessments (Mondaq/DGL). The legislature's bill page (primary) records it becoming law on 14 May 2026, developer duties from 1 January 2027, and post-adverse-outcome disclosure rules that the Attorney General must adopt by that date (Colorado General Assembly). The Attorney General's draft rules, which also cover the Chatbot Safety Act, are open for comment until the hearing on 26 October 2026 (Colorado Attorney General). xAI sued Colorado in X.AI LLC v. Weiser (D. Colo., No. 1:26-cv-01515), filed 9 April 2026 against the Attorney General over SB 24-205. The United States intervened as a plaintiff on 24 April 2026. A minute order of 27 April 2026 (ECF 24) bars the AG from beginning enforcement for alleged violations of SB 24-205, or of legislation replacing or amending it enacted that session, until 14 days after a ruling on xAI's preliminary-injunction motion, which is due within 28 days after final implementing rules; other deadlines are stayed. Whether the stay reaches SB26-189 is arguable from the order's wording, not confirmed (primary RECAP docket, entries to 27 April 2026) (CourtListener: X.AI LLC v. Weiser).

California's SB 53 and New York's RAISE Act regulate frontier developers. A business that builds on a frontier API is unlikely to be covered by either. California's ADMT rules and Colorado's law, by contrast, reach businesses that deploy automated decisions, which is where most business automation lands. RAISE remains unsettled: secondary sources report that its chapter amendments (A9449/S8828) passed on 11 March and were signed on 27 March 2026, effective 1 January 2027, while Wikipedia (cached February 2026) still shows them in committee; New York's official sites could not be reached (SecurePrivacy; Wikipedia: RAISE Act). State attorneys general also act without AI statutes. California's AG opened an investigation into Grok's sexualised images in January 2026, and 35 state AGs wrote to xAI (secondary) (Wikipedia: Grok sexual deepfake scandal). Florida sued OpenAI on 1 June 2026 (secondary) (Wikipedia: 2026 in artificial intelligence). xAI (SpaceXAI LLC) is also seeking an injunction pending appeal against Minnesota's Attorney General in the Eighth Circuit (No. 26-2806; opposition filed 21 Sep 2026; the district-court ruling was not read) (CourtListener: SpaceXAI LLC v. Ellison). For hiring tools, New York City's Local Law 144 has been enforced since 5 July 2023: an automated employment decision tool needs a bias audit within the year before use, a public audit summary and notice to candidates (NYC DCWP). Illinois's employment AI law was not verified.

Adoption: half of adults, one firm in five

The Census Bureau's Business Trends and Outlook Survey found 17-20% of firms using AI between December 2025 and May 2026, with 20-23% expecting to adopt within six months. Adoption reached 37% among firms with 250 or more employees, against under 20% for firms with four or fewer (US Census Bureau). The question was widened in November 2025, so earlier figures do not compare cleanly (Federal Reserve FEDS Note). Later secondary estimates of 21-22% are unchecked against Census tables (Keyhole Software). Stanford's 2026 AI Index gives the US a generative-AI adoption figure of 28.3% (24th), but it is not a measure of firms: the page does not call it organisational adoption, and the value and rank match Microsoft's population figure for the second half of 2025 (Stanford HAI, Economy; Microsoft AI Diffusion Report Q1 2026).

Individuals are well ahead of their employers. Pew's February 2026 survey (n=5,119) found that 49% of US adults use AI chatbots, up from 33% in 2024, and 24% use them daily. Part of the rise reflects a sampling change, since the 2024 question went only to adults who had heard of chatbots (Pew; Pew, daily use). Gallup's May 2026 indicator shows 52% of employees using AI at work, 15% of them daily (Gallup). Anthropic's per-capita Claude usage index for the US was 4.25 in April and 3.87 in May 2026 (1.0 = proportional to working-age population; consumer plans only) (Anthropic Economic Index data). Attitudes are the weak point. Roughly two-thirds of Americans say AI is moving too fast; summaries give 63% or 67% (TheNextWeb). Only 31% trust their government to regulate AI, the lowest share Stanford surveyed, against a 54% global average (Stanford HAI, Public Opinion).

The IMF puts almost 60% of US employment in highly exposed occupations, which measures task overlap, not job loss (IMF Staff Discussion Note). Early effects are concentrated among young workers. Employment of workers aged 22-25 in AI-exposed occupations is 19% below the path it would have followed alongside less-exposed peers, through reduced hiring, with no comparable gap for experienced workers (Stanford Digital Economy Lab, revised 12 August 2026; ADP payroll data to June 2026). The authors call these "early, descriptive indicators ... rather than causal estimates"; the gap attenuates when education is controlled for and is stronger in ADP than in national surveys (Stanford Digital Economy Lab: Canaries in the Coal Mine).

Compute and capital: the gigawatt economy

SEC filings give the hard numbers on capex:

• Microsoft: $115.95 billion in fiscal 2026, against $64.55 billion the year before (SEC XBRL, Microsoft).
• Alphabet: $80.60 billion in H1 2026, against $39.64 billion in H1 2025 (SEC XBRL, Alphabet).
• Meta: $49.11 billion in H1 2026, against $29.48 billion (SEC XBRL, Meta).
• Amazon: $98.41 billion in H1 2026, against $57.20 billion. This figure uses a non-standard XBRL tag and should be checked against the cash-flow statement (SEC XBRL, Amazon).

First-quarter guidance implies about $700-725 billion for 2026 across the four companies (secondary; Meta's range varies by source) (Yahoo Finance). Revenue lags the spending. Microsoft's AI business ran at $37 billion a year in the March 2026 quarter (Microsoft FY26 Q3), and Nvidia's data-centre revenue reached $89.0 billion in the quarter to 26 July 2026 (total revenue about $96.2 billion; results of 26 August 2026; secondary, the release was not re-read on 12 October 2026) (NVIDIA Q2 FY2027). Epoch AI says hyperscaler capex "is on trend to outpace their cash inflows by the end of 2026" (Epoch AI).

The Stargate venture plans up to $500 billion of US AI infrastructure by 2029 (Wikipedia: Stargate LLC). OpenAI said in late April 2026 that it had 10 GW under contract, which is not the same as 10 GW operating (Bloomberg Law, Stargate). Epoch AI tracks seven US sites. The flagship at Abilene, Texas, is designed for 1.2 GW by Q4 2026, but its operating capacity is unresolved: two versions of Epoch's page say 0.6 GW and 0.3 GW (Epoch AI, Stargate sites). In August 2026 the Nvidia arrangement was amended to an 8 GW leased site in Pike County, Ohio, with a $105 billion Nvidia backstop (secondary) (Wikipedia: OpenAI).

The US holds 25 of the TOP500 top 100 systems, although rank 1 in June 2026 is Chinese (TOP500, June 2026). Stanford says US private AI investment was 23 times China's, and that in generative AI US investment "exceeded the combined total of China and Europe by a wide margin" (Stanford HAI, Economy). The report PDF gives US private AI investment of $285.9 billion in 2025, 23.1 times China's $12.4 billion and 48.5 times the UK's $5.9 billion; in generative AI the US had $163.6 billion against $1.48 billion for China and $3.21 billion for Europe (Quid data; funding rounds, not spending) (Stanford AI Index 2026 PDF). See Costs and Hardware for more.

Canada: a sovereign-compute bet anchored on one company

Statistics Canada's Q2 2026 survey found 19.2% of businesses using AI to produce goods or deliver services in the previous 12 months (Statistics Canada, checked 11 Oct 2026). Its analysis of the same survey (11 June 2026) shows that the share has tripled from 6.1% in Q2 2024, but also a long tail: 40.0% of businesses said AI was not relevant to them. Information and cultural industries (42.3%), finance and insurance (40.4%) and professional, scientific and technical services (32.4%) lead adoption, and 27.8% of businesses with 100 or more employees use AI (Statistics Canada, AI use analysis, checked 11 Oct 2026). Individual use, at 38.5%, is well above the US level (Microsoft Q2 2026), and Anthropic's usage index scores Canada at 2.91 (Anthropic Economic Index, Sept 2025).

The federal strategy is built around compute. The C$2 billion Sovereign AI Compute Strategy comprises up to C$700 million for private compute, C$1 billion for public infrastructure and a C$300 million access fund. Its first allocation was up to C$240 million to Cohere for a data-centre project of about C$725 million in Cambridge, Ontario; whether it has opened was not found (secondary) (The Logic). That makes Cohere the national champion and a proxy for the programme's progress. In September 2026 Cohere was reported to be in advanced, unfinalised talks to raise US$2-3 billion at a US$20 billion valuation (Forkast, citing the Globe and Mail). On 16 September it signed a definitive agreement to combine with Germany's Aleph Alpha. The combined company will have more than 1,000 staff and headquarters in both Berlin and Toronto, subject to regulatory approval (Cohere blog).

Canada has no AI statute. The proposed Artificial Intelligence and Data Act (AIDA) was part of Bill C-27, which was still in committee when the parliamentary session ended on 6 January 2025 (LEGISinfo, C-27, checked 11 Oct 2026). The current national strategy, "AI for All", covers six areas from protecting Canadians to sovereign infrastructure and points to privacy and online-safety laws as the vehicles for trust; it names no successor AI bill (ISED, National AI Strategy). Evan Solomon has been minister of artificial intelligence and digital innovation since May 2025 (secondary) (Wikipedia: Evan Solomon). Enforcement is coming through other channels. In June 2026 the federal Privacy Commissioner found that Grok's image generation broke Canadian privacy law (secondary) (Wikipedia: Grok sexual deepfake scandal). In September, British Columbia sued OpenAI and Sam Altman over the Tumbler Ridge school shooting (The Guardian).

Latin America: plans and bills, few binding rules

The region's large economies sit near the global average for individual use: 20.2% in Brazil and 21.4% in Mexico, against 18.8% worldwide. The smaller ones score higher: Colombia 25.9%, Chile 24.0% and Argentina 23.4% (Microsoft Q2 2026, checked 11 Oct 2026). On preparedness, the IMF scores Chile 0.586, Mexico 0.532, Brazil 0.501, Colombia 0.489 and Argentina 0.474 (IMF DataMapper, checked 11 Oct 2026). The research found no firm-level adoption data for any Latin American country.

Brazil

Brazil's AI bill, PL 2338/2023, is Latin America's most developed, and it has stalled. It passed the Senate on 10 December 2024 and was presented to the Chamber of Deputies on 17 March 2025 (Senado Federal, PL 2338/2023). The Chamber's record shows it awaiting the rapporteur's opinion in a special committee, with no floor vote; the latest entries are dated 1 and 2 September 2026 (Câmara dos Deputados, PL 2338/2023). The rapporteur is Aguinaldo Ribeiro (secondary) (CASRAI: Brazil AI bill status). Planned votes in February and then the first half of 2026 did not happen (Teletime). In August, Ribeiro said a vote would probably wait until after the October 2026 elections (MobileTime, 24 Aug 2026; not re-read on 12 October 2026; the Chamber record sets no vote date).

The national AI plan (PBIA) has a widely reported baseline of R$23 billion for 2024-2028. Of that, R$18.2 billion comes from the FNDCT science fund, R$2.9 billion from the federal budget, R$0.43 billion from state-owned companies and about R$1 billion from private partners (single source for the split) (Times Brasil). A single August 2026 report describes a revised plan with more than R$2 billion for two supercomputers, one American and one Chinese, plus a Rio data centre with Chinese partners (Paraíba Business). The research gathered no figures on Brazilian enterprise adoption or government use.

Mexico

Mexico's federal digital agency (ATDT) published a National AI Plan 2026 in September. It rests on three pillars: an ethical and legal framework, talent and public software, and sovereign infrastructure. The plan proposes a federal legal framework. Its flagship is the Coatlicue supercomputer, built with SECIHTI, at about MXN 6,000 million and a target of 314 petaflops. These details come from a single news article; the plan text itself was not read (IDC). A single secondary source says Mexico had no general AI law as of July 2026 and that a Senate base text was not voted before the April 2026 session ended. The same source counts 85 AI initiatives since April 2023, 67 of them pending (Ecosistema Startup).

Chile, Colombia and Argentina

Chile's AI bill (merged Boletines 15.869-19 and 16.821-19) passed the Chamber of Deputies on 13 October 2025 and went to the Senate. No Senate action in 2026 was found (IAPP).

Colombia's policy is CONPES 4144, dated 14 February 2025 and covering 2024-2030. Its six axes are ethics and governance; data and infrastructure; research and innovation; talent; risk mitigation; and adoption in public bodies, businesses and regions. The approved PDF ("Versión aprobada", 14 February 2025) gives 106 actions to 2030 costing about COP 479,273 million; the digits are reconstructed from a split PDF text layer (primary) (DNP, CONPES 4144).

For Argentina, the only policy-related data point is that OpenAI's October 2025 letter of intent for a Stargate Argentina with Sur Energy had not progressed by May 2026 (secondary) (Wikipedia: Stargate LLC). Beyond Microsoft's usage figures, data are thin across all three countries.

European Union: one rulebook, delayed high-risk dates, public money for compute

The AI Act timetable after the Omnibus

The Commission proposed the AI Omnibus on 19 November 2025. A political agreement followed on 7 May 2026, and the Omnibus entered into force on 27 July 2026. The Commission says the AI Act "became applicable on 2 August 2026, with some exceptions" (European Commission, AI Act, checked 11 Oct 2026).

Sources: (European Commission; AI Act Service Desk, implementation timeline; European Commission, transparency enforcement press release; Addleshaw Goddard; NatLawReview; Sidley; Commission pages checked 11 Oct 2026). The Commission links the final text as Regulation (EU) 2026/1744 (primary); its Official Journal publication on 24 July 2026 rests on law firms (NatLawReview), EUR-Lex did not load, and sources disagree on the Council's approval date.

The Omnibus also dropped a mechanism that would have tied the high-risk deadlines to the availability of standards, so the new dates are fixed (Addleshaw Goddard). When the proposal was made in November 2025, Executive Vice-President Henna Virkkunen said: "This is not about rolling back the rules" (Mobile World Live). Civil-society groups replied that people would go unprotected for more than a year (secondary) (CASRAI: AI Act deadline delay).

For businesses that build or deploy automation, the live duties are the transparency ones. Article 50(1) requires that people be told they are interacting with an AI system unless that is obvious. Article 50(4) requires disclosure of AI-generated text on matters of public interest unless it has had human editorial control (secondary) (artificialintelligenceact.eu, Art. 50). Since 2 August 2026, deepfakes must also be labelled and AI-generated content must carry machine-readable marks (European Commission, press release). So a chat or voice agent deployed in the EU needs disclosure now, while a hiring-screening tool has until December 2027 to meet the high-risk rules. The AI-literacy duty also applies now. After the Omnibus it sets no specific "sufficient" level, needs no certificate, and can be evidenced by an internal training record; national authorities enforce it from August 2026 (European Commission, AI literacy Q&A). Penalties (Art. 99): prohibited practices up to €35 million or 7% of worldwide annual turnover; most other operator duties €15 million or 3%; incorrect or misleading information €7.5 million or 1% (whichever is higher; for SMEs, lower). GPAI providers (Art. 101): up to €15 million or 3%, enforced by the AI Office since 2 August 2026 (Commission, primary). The penalty text was read from a reproduction (secondary; EUR-Lex unreadable) (artificialintelligenceact.eu, Art. 99).

Rules beyond the AI Act that bite on automation

Three other instruments matter as much as the AI Act for automation work:

• GDPR Article 22. People have the right not to be subject to solely automated decisions with legal or similarly significant effects. Where an exception applies, they keep the right to human intervention, to express a view and to contest the decision (GDPR Art. 22).
• Digital Services Act. The DSA reaches platforms that deploy AI features. On 26 January 2026 the Commission opened a formal proceeding against X over Grok, with fines of up to 6% of global turnover possible (Anadolu Agency).
• GPAI Code of Practice (10 July 2025). This is the practical signal for model choice. The Commission's page confirms that xAI "signed up to the Safety and Security Chapter" only, so it must show compliance with the transparency and copyright duties by other means. A summarising fetch of the same page listed 22 signatories (among them Amazon, Anthropic, Cohere, Google, IBM, Microsoft, Mistral AI, OpenAI and ServiceNow); re-check the names on the page before relying on them (secondary). Meta does not appear on the Commission's signatory page (12 Oct 2026) (European Commission, GPAI Code).

Investment and compute: gigafactories and an unverified EUR 200 billion

EuroHPC's AI Gigafactories call opened on 30 July 2026 and its submission deadline is 12 November 2026. It will select up to seven gigafactories in at least seven member states in early 2027, to begin operating within 18 months. EU and national funding acts as anchor customer and is expected to unlock more than EUR 20 billion of private investment; the bulk of the public money is to come from the next EU budget (2028-2035) (EuroHPC JU, checked 11 Oct 2026). Secondary reports put the public share at EUR 10 billion and describe letters of intent from AMD, Nvidia and Qualcomm on chip access (StorageNewsletter). The "InvestAI" headline of EUR 200 billion could not be verified, because the Commission page returned an error. The scale gap with the US is stark: roughly EUR 30 billion across up to seven sites over several years, against $80.6 billion of capex by Alphabet alone in H1 2026.

Adoption: an eightfold spread from Denmark to Romania

In 2025, 19.95% of EU enterprises with ten or more staff used AI, up from 13.48% in 2024 and 7.65% in 2021. Size matters: 55.03% of large firms (250+) used AI, against 18.90% of SMEs (10-249). The euro-area figure was 21.37% (Eurostat API; WAM, citing Eurostat). The commonest use is analysing written language (11.8% of enterprises). About 62.5% of information and communication firms use AI, and about 40.4% of professional and technical firms; every other sector is below 25% (secondary) (WAM). Denmark (+14.5 points), Finland (+13.5) and Lithuania (+12.5) grew fastest, while Greece slipped slightly (8.93% against 9.81%) (Eurostat API; Implicator.ai).

Source: (Eurostat API, isoc_eb_ai, 2025; % of enterprises using at least one AI technology; all values re-read on 12 Oct 2026, dataset updated 15 Jun 2026). The dataset has no 2022 or 2026 values, and Iceland, Switzerland and the UK are not in it. Romania's rate is about one-eighth of Denmark's. EU large firms are almost three times as likely as SMEs to use AI, and the gap is widest where adoption is lowest: in Poland, 45.81% of large firms against 7.09% of SMEs. Eurostat flags a break in the Netherlands' series, so read its change on 2024 with care. Early press reports of 42.0% for Sweden were a misreading.

European companies to know

Mistral AI (France) raised EUR 1.7 billion in September 2025, then a EUR 3 billion Series D at more than EUR 21 billion post-money on 8 September 2026, pitching "sovereign, open-weight AI" (Mistral news). Mistral Large 4 (6 October 2026) has 1 trillion parameters, 52 billion of them active. It was trained on 3,800 Nvidia Grace Blackwell GPUs in Mistral's own European data centres, and open weights are promised by the end of October (Mistral, Large 4). The company has 2026 partnerships with Cloudera, Mozilla, Nvidia and Saudi Arabia's HUMAIN, and opened a Munich hub on 28 September. Its licensing needs checking model by model: its FAQ says commercial deployment of open-weight models needs a Mistral licence, while the model cards carry Apache 2.0 tags (Mistral pricing).

Other European signals:

• The Cohere-Aleph Alpha combination (see Canada).
• Yann LeCun's Paris-based AMI Labs, which raised US$1.03 billion in March 2026 (Sifted).
• SAP's May 2026 investment in the workflow platform n8n at a US$5.2 billion valuation, with n8n to be embedded in SAP's Joule Studio (n8n press).

Stanford's AI Index 2026 gives Europe (region) $20.92 billion of private AI investment in 2025 and $3.21 billion in generative AI; cumulative 2013-2025 investment was $34.1 billion for the UK and $17.2 billion for Germany, and European public AI contracts in 2013-2024 totalled about $3.7 billion (UK $1.6 billion, Germany $505 million, France $320 million) (primary PDF, Quid data) (Stanford AI Index 2026 PDF). See Companies and Other models.

European countries

United Kingdom: new departments, Growth Zones and a permissive data law

Government. gov.uk says DSIT "is changing" and is being replaced by the Department for Business, Innovation, Science and Trade (DBIST), DCMS and the Cabinet Office (gov.uk, DSIT). On 20 July 2026 Kanishka Narayan was appointed Minister of State (Minister for Artificial Intelligence), working jointly in the Cabinet Office and DBIST and attending Cabinet; the same appointments named Jonathan Reynolds Secretary of State for Business, Innovation, Science and Trade (gov.uk, Ministerial appointments; both GOV.UK pages re-read on 12 Oct 2026, primary). Andy Burnham has been Prime Minister since 20 July 2026, succeeding Keir Starmer (Prime Minister from 5 July 2024) (gov.uk, Andy Burnham, primary). Two points remain unverified: the formal date and cause of DSIT's replacement, and a trade-press claim that Growth Zone work moved departments. The AI Security Institute is led by Director Henry de Zoete. Jade Leung is CTO and also the Prime Minister's AI Adviser, Ian Hogarth is Chair and Chris Summerfield is Research Director (AISI). According to Wikipedia, the institute had pre-release testing agreements with Anthropic, Google and OpenAI as of February 2026 (secondary) (Wikipedia: AI Security Institute).

Compute. Five AI Growth Zones have been announced: Oxfordshire/Culham (April 2025), North East England (September 2025), North Wales/Anglesey (November 2025), South Wales (Newport to Bridgend, November 2025) and Lanarkshire (January 2026) (gov.uk, AI Growth Zones). They have more than GBP 28 billion of committed private investment (UK Parliament, May 2026; not re-read on 12 October 2026, page returned 403). Each zone gets GBP 5 million for business adoption and skills; South Wales is linked to GBP 10 billion and "over 1GW by early 2030s". GOV.UK separately cites GBP 24.25 billion of private investment committed across all sectors in the month to 20 November 2025, and says the reforms could "unlock up to £100 billion" (primary) (gov.uk, 20 Nov 2025). A target of 6 GW of new AI-ready capacity by 2030 appears only in secondary sources and is unverified (CASRAI: AI Growth Zones). There are warning signs: OpenAI paused Stargate UK in about April 2026, citing energy costs and regulatory uncertainty, and says it could resume (secondary, several outlets) (TechRadar). The government's "One Year On" report (29 January 2026) says commitments have been met on 38 of the plan's 50 actions (GOV.UK, AI Opportunities Action Plan: One Year On, primary).

Rules. The research found no UK AI statute. The Data (Use and Access) Act 2025 creates "a more permissive framework" for solely automated decisions, subject to safeguards (gov.uk, DUAA). Section 80, which inserts UK GDPR Articles 22A-22D, has been in force since 5 February 2026 under S.I. 2026/82, reg. 2(j) (legislation.gov.uk, primary). Other bodies are also active. Ofcom opened a Grok investigation on 12 January 2026, and the UK criminalised creating non-consensual intimate AI images (secondary) (Wikipedia: Grok sexual deepfake scandal). In November 2025 the High Court found that Stability AI had not infringed copyright by training on Getty images (secondary) (Wikipedia: Stability AI). Like the US, the UK declined to sign the Paris AI Action Summit declaration (secondary) (Wikipedia: 2025 in AI).

Use. ONS found 25% of UK businesses using some AI in late December 2025, up 15 points since late September 2023 (ONS, 8 Jan 2026, primary); a June 2026 figure of 29% appears only in news coverage (single-source). Government has measured its own use: a cross-government Microsoft 365 Copilot experiment (30 Sep-31 Dec 2024; about 20,000 staff in 12 organisations) found a self-reported saving of 26 minutes a day, and 82% would not want to go back; an HMRC phase reported 2-3% of the working week (about 60 minutes), adjusted down about 20% as a baseline estimate (GOV.UK, Copilot cross-government findings, primary; self-reported). The other examples are vendor cases. Anthropic says Barclays' retrieval-based Colleague Knowledge Assistant is used by more than 16,000 staff and that the bank targets Claude Code use by 50% of its developers by end-2026, with no savings published (Anthropic, Barclays). The retailer Toolstation cut zero-result searches from 2% to 0.1% and lifted search revenue 5.5% with Gemini-based search (Google Cloud). The IMF exposure figure in the table comes from its 2024 note (IMF).

France: heavy individual use, foreign-funded data centres, Europe's model maker

France has the widest gap in this report, in percentage points, between individual and firm use (Poland has the highest ratio; see below). Almost half of working-age people use generative AI, but only 18.16% of enterprises do, below the EU average (Microsoft Q2 2026; Eurostat; IMF DataMapper; TOP500).

The headline investment is EUR 109 billion, announced at the February 2025 summit. It is mostly private and spread over several years (SiliconANGLE). Named commitments include EUR 30-50 billion from the UAE, through MGX, and EUR 20 billion from Brookfield for a data centre at Cambrai; outlets differ on the figures (The Media Leader). Delivery has not been verified, so the figure should be read as investor intent.

France's strength is its model companies, Mistral and AMI Labs (see the EU section). In testimony to the National Assembly on 12 May 2026, Mistral CEO Arthur Mensch did not rule out rising unemployment in some sectors and flagged electricity demand as an inflation risk (LCP). Paris prosecutors searched X's offices with Europol on 3 February 2026 as part of the Grok inquiry (secondary) (Wikipedia: Grok sexual deepfake scandal). The national strategy, the AI Act authority and public-sector use were not researched.

Germany: mid-table individual use, strong firms and compute, a Canadian merger

Germany is France's mirror image: individual use is lower than in the US, but firm adoption is well above the EU average, and it has more top-100 supercomputers than any other European country (Microsoft Q2 2026; Eurostat; TOP500). Aleph Alpha's combination with Cohere gives the merged company a Berlin headquarters (Cohere blog), and Mistral's new Munich hub targets German industry. The federal strategy site's newest documents are the 2020 update and the 2023 Aktionsplan KI (primary) (KI-Strategie Deutschland); the federal research ministry's AI pages, gigafactory bids, compute and Destatis data were not read (BMFTR). A secondary ranking puts Germany ninth in the Tortoise Global AI Index (Comland).

Nordics: Europe's adoption leaders, policy unresearched

Sources: (Eurostat; Implicator.ai; Microsoft Q2 2026; IMF DataMapper; TOP500).

Denmark and Finland lead Europe on business adoption and also grew fastest in 2025, so diffusion is accelerating where it was already high. They are the only countries in this report where more firms use AI than individuals do: in Denmark 42.03% of enterprises against 32.4% of working-age people, in Finland 37.82% against 30.6%. Denmark also has the highest IMF preparedness score here (0.779). Norway shows the opposite gap: individual use is very high, at 49.4% (third in Europe), while firm adoption trails its neighbours. The only compute item is that Microsoft took over OpenAI's Stargate Norway capacity in April 2026 (headline only) (Hacker News listing). The research covered no national strategies, regulators or public-sector use for any of the four countries.

Netherlands and Benelux: high on every adoption measure

The Netherlands combines 43.5% individual use, 33.21% firm adoption and an IMF score of 0.766. Among the IMF values in this report only Denmark (0.779) and the US (0.771) score higher (IMF DataMapper). Belgium and Luxembourg are close behind on firm adoption. The Netherlands ranks 13th in the Tortoise index (secondary) (Comland). Dutch strategy and regulators were not researched.

Italy: the only national AI statute found in this research

Law No. 132 of 23 September 2025 sets out provisions on AI and delegates powers to the government. It was published in the Gazzetta Ufficiale on 25 September and entered into force on 10 October 2025. It is the only national AI statute in the EU found by this research (Gazzetta Ufficiale, checked 11 Oct 2026). Its delegated decrees and supervisory roles were not read, so businesses operating in Italy or serving Italian customers should check them with local counsel. Italy is below the EU average on adoption but strong on supercomputing: five of the June 2026 top 100, and two of the global top 10 in November 2025, second only to the US (TOP500, November 2025).

Spain, Poland, Estonia, Ireland and Switzerland: adoption data only

These five profiles rest mainly on adoption figures, and data are thin for each.

• Spain: fourth in Europe on individual use, but firm adoption is only at the EU average. The status of Spain's AI agency (AESIA) could not be verified because its site failed to load.
• Poland: the second-lowest firm adoption in the EU after Romania (5.21%), just below Bulgaria. Yet individual use, at 32.3%, matches Germany's, so Poland has the highest ratio of individual to firm use in this report, almost four to one. The research found nothing else.
• Estonia: firm adoption above the EU's 19.95% and well above Latvia's 12.21%, and an IMF score (0.764) close to the Netherlands'. No policy data were found.
• Ireland: the highest European individual-use figure, but firm adoption just below the EU average. Ireland is an active enforcer. Its Data Protection Commission opened a GDPR inquiry into X over Grok in February 2026, the media regulator Coimisiún na Meán consulted the Commission, and police reported 244 related investigations by March (secondary) (Wikipedia: Grok sexual deepfake scandal).
• Switzerland: its IMF score (0.757) is among the highest here, after Denmark, the US, the Netherlands, Estonia and Finland, and it ranks 12th in the Tortoise index (secondary) (Comland).

Sources for the figures above: (Microsoft Q2 2026; Eurostat; IMF DataMapper; TOP500).

Cross-country themes

The adoption gap is a deployment gap, not an awareness gap. In almost every country with both measures, more individuals use AI than firms do, and in several the ratio exceeds two to one:

The gap is narrowest where firm adoption is highest. Sweden (37.3% against 35.04%), Belgium, the Netherlands and Germany show that it can close, and in Denmark and Finland firms are ahead of individuals. The measures differ, so the comparison is indicative, but the pattern is consistent. Firm size matters too: in the US, 37% of firms with 250 or more staff use AI, against under 20% of the smallest (US Census Bureau); in the EU, 55.03% of large firms (250+) do, against 18.90% of SMEs (Eurostat API); and in the UK, 44% of businesses with 250 or more employees, against 25% overall (ONS).

Three regulatory models are converging on the same controls. The EU has one binding, risk-tiered law with fixed dates. The US federal government favours voluntary standards and works against state rules, while the states pass narrower laws. After Colorado's rewrite, those laws centre on notice, explanation of adverse outcomes and human review. The UK, Canada and Latin America rely mainly on data-protection law and sector regulators, and their comprehensive bills have stalled or lapsed. In practice the controls converge. GDPR Article 22, Colorado SB26-189, California's ADMT rules and the UK DUAA safeguards all ask for the same three things: tell people when AI is involved, explain adverse automated decisions, and offer human review.

Compute is the axis of sovereignty. The US builds at hyperscaler scale with private capital. Europe pools public money and backs champions that sell sovereignty, such as Mistral's own data centres and the Cohere-Aleph Alpha combination. Middle powers buy national flagships: Canada's C$2 billion programme, Mexico's Coatlicue, and Brazil's planned supercomputers. Gulf partners appear through the UAE commitments in France's package and Mistral's HUMAIN partnership, whose terms were not disclosed. See Middle East and Africa and Future outlook.

Model access is now a policy risk. The June 2026 directive took two frontier models offline worldwide for 19 days. Add the reported vetting of GPT-5.6 Sol users and the Pentagon's action against Anthropic, and US model availability can change for reasons unrelated to price or performance (Anthropic; BBC). European organisations that weigh sovereignty are likely to cite these events when they choose models. See Claude, OpenAI and the timeline.

Existing law is doing the enforcing. Within six months of the Grok deepfake case, six jurisdictions covered here had responded: a Digital Services Act proceeding (EU), a GDPR inquiry (Ireland), an Ofcom investigation (UK), a prosecutors' search of X's offices (France), a Privacy Commissioner finding (Canada), and a California attorney-general investigation plus a letter from 35 state attorneys general (US) (Wikipedia: Grok sexual deepfake scandal; Anadolu Agency). None of these actions needed an AI statute. See Grok. For businesses that deploy automation, privacy, consumer-protection and online-safety law are the near-term enforcement risk.

Trust is lowest where AI investment is highest. Americans are the least likely of any surveyed population to trust their government to regulate AI (31%), and about two-thirds say AI is moving too fast (Stanford HAI, Public Opinion). That helps explain why US states keep legislating despite federal pressure, and why disclosure rules survive where comprehensive laws fail. Concern is rising in Europe too: in Pew's 2026 survey (42,151 adults; 37 countries and territories listed), "more concerned than excited" rose by 9 points in Sweden, 8 in Poland, 7 in the Netherlands and 6 in Hungary (Pew, 17 Sep 2026). The latest Eurobarometer on AI (February 2025) found more than 60% of Europeans positive about AI at work and 84% wanting careful management for privacy and transparency (secondary) (European Commission, 13 Feb 2025).

Compliance checklist for US and EU deployments

Use these questions to settle scope with your vendor or implementation partner and with counsel. This is not legal advice; who counts as "developer", "deployer" or "provider" depends on contracts and facts. Security controls (the OWASP lists and NIST) are covered in Regulation and security.

United States

• Map your state footprint. Locate the people affected by your automation, not just your place of incorporation.
• Texas (since 1 Jan 2026, per earlier sources). Confirm that no system is designed to incite harm, discriminate unlawfully or create prohibited deepfakes. Document NIST AI RMF alignment, which supports the statutory safe harbour.
• Colorado SB26-189 (from 1 Jan 2027). If you build an ADMT used for consequential decisions, give the deployer technical documentation (intended uses, training-data categories, known limitations, instructions for use and human review); if you buy one, ask the developer for it. Deployers need point-of-interaction notice, a plain-language explanation within 30 days of an adverse outcome, and human review on request; both keep records for 3 years. The AG's draft rules are open for comment until 26 Oct 2026, with final rules due by 1 Jan 2027. Track X.AI v. Weiser: the AG cannot begin enforcement until 14 days after a preliminary-injunction ruling (whether this covers SB26-189 is arguable). Build to the 1 Jan 2027 duties anyway.
• Colorado Chatbot Safety Act, HB 26-1263 (from 1 Jan 2027). Chatbot operators must disclose that users are talking to AI, estimate user age, protect teen users and have self-harm response protocols.
• California ADMT rules (duties from 1 Jan 2027; risk assessments since 1 Jan 2026). Identify any significant decisions your automation supports and your deadlines: ADMT compliance by 1 Jan 2027, risk-assessment submissions by 1 Apr 2028 (for 2026-2027 assessments), and the first cybersecurity audit on 1 Apr 2028, 2029 or 2030 depending on revenue.
• California SB 53 and New York RAISE. These target frontier developers. Confirm your business is not one, and track the RAISE chapter amendments (reported signed 27 Mar 2026 and effective 1 Jan 2027; secondary, conflicting with Wikipedia).
• Federal preemption. Do not plan around it. EO 14365 has no court ruling behind it, so build for state law as written.
• Hiring tools. NYC Local Law 144 (enforced since 5 Jul 2023) requires a bias audit within the year before use, a public audit summary, and notice to candidates 10 business days before use (NYC DCWP). Check Illinois's employment AI law separately; it was not verified here.
• Model continuity. Keep a tested fallback model. Federal contractors should also check whether their model vendor faces procurement restrictions.
• Vendor indemnities. Read each one closely. Anthropic's covers third-party IP claims over paid use and outputs, but excludes customer inputs and combinations with other technology (Anthropic Commercial Terms). Those exclusions cover exactly the parts you or your integrator add.

European Union and UK

• Disclosure (since 2 Aug 2026). Chat and voice agents must tell people they are talking to AI unless that is obvious (Art. 50(1)). Deepfakes must be labelled, and AI-generated text on public-interest matters needs disclosure unless it has had human editorial review (Art. 50(4)).
• Content marking (since 2 Aug 2026). AI-generated content needs machine-readable marks (Art. 50(2)). Only generative systems already on the market before 2 Aug 2026 have a grace period, to 2 Dec 2026.
• Prohibited uses (from 2 Dec 2026). Exclude anything touching the new bans on AI systems that generate non-consensual sexual deepfakes or child sexual abuse material.
• AI literacy (since 2 Feb 2025; enforced nationally from Aug 2026). Keep an internal record of role-based AI training for staff who build or operate the system; no certificate is required (Art. 4).
• High-risk uses (Annex III, from 2 Dec 2027). For hiring, credit scoring and the other Annex III uses, start documentation now. Product-embedded AI (Annex I) follows on 2 Aug 2028. Settle the provider and deployer roles with counsel.
• GDPR Article 22. Solely automated decisions with significant effects need a lawful exception, plus human intervention, the right to express a view and the right to contest.
• Model selection. Prefer GPAI Code signatories and record the choice. Note that xAI signed only one chapter.
• Data residency. Confirm the processing region. Anthropic's own API offers only "global" or "us" inference (inference_geo; US-only costs 1.1x), and the setting does not apply on Bedrock, Google Cloud or Foundry (Anthropic, data residency); its compliance page listed regional residency through cloud partners (read 11 Oct 2026; not re-readable 12 Oct) (Claude regional compliance). OpenAI processes in-region only in the US, the EU and the UAE (OpenAI, your data); Mistral has EU and US endpoints (secondary). Confirm the region on each cloud's model pages.
• National layers. Check each country where you operate or have customers for its own AI law, for example Italy's Law 132/2025.
• UK. Apply the DUAA safeguards for automated decisions, and check Online Safety Act exposure for any image or user-content features.

What this means for your business

Start where your staff are ahead of your processes. Across the US, Canada and the EU, about one firm in five uses AI (one in four in the UK), while a third to a half of working-age people already do. If your staff already use chatbots but your processes do not, begin by turning that informal use into managed, logged workflows. Information and professional services lead firm adoption in Canada and the EU, as do finance and insurance in Canada, so peers there offer the most evidence. See Use cases and the Playbook.

Build the compliance components once, in the first release. The EU transparency duties have applied since August 2026, and several deadlines cluster: Colorado's ADMT and chatbot laws and California's ADMT rules on 1 January 2027 (a federal court order holds back Colorado's enforcement until after a preliminary-injunction ruling, but the duties stand), and the EU milestones in December 2026 and December 2027. Because the controls converge, one set of notice templates, adverse-outcome explanations, human-review queues, audit logs and AI-disclosure banners can cover several US states and the EU. Budget for them up front, not as a retrofit; the checklist above shows which apply.

Plan for a model going offline. The June 2026 shutdown showed that a single-vendor architecture can fail for reasons outside any contract. Require a model-agnostic design and test the fallback before go-live. For EU deployments, GPAI Code signatories and EU-region processing reduce friction; the European-headquartered options are Mistral and, once it closes, the Cohere-Aleph Alpha combination (dual headquarters in Berlin and Toronto). US-only processing on Anthropic's API costs 1.1x. See Top models, Costs and the Tool stack.

Expect model vendors among your implementation options. OpenAI launched a deployment and integration company, DeployCo, in May 2026 with more than US$4 billion of initial investment (secondary) (Wikipedia: OpenAI). Anthropic committed US$100 million to train 10,000 "Frontier Deployed Engineers" by the end of 2027, with Accenture, Deloitte, McKinsey and Capgemini in the early cohort (Anthropic, Frontier Academy). Whoever you compare, ask about their sector experience and the certifications you require. If you run SAP, note SAP's n8n investment (n8n press).

In Latin America, plan without fixed dates. Brazil's bill is still awaiting the rapporteur's opinion in a Chamber special committee; the rapporteur expected a vote only after the October 2026 elections (MobileTime, 24 Aug 2026; not re-read on 12 October 2026), and its application dates are unknown. Build to data-protection law and the common controls above, and track the bill. If your vendor processes data in India, its DPDP obligations are covered in India.

Questions to ask any vendor or implementation partner: Which US state and EU Article 50 duties apply, and who produces the notices and records? Who is "developer", "deployer" or "provider", and does the contract say so? When was the fallback model last tested? In which region and cloud is the data processed? What does the model vendor's indemnity exclude, and who covers those parts?

Open questions and unverified items

• New York RAISE amendments: reported signed 27 March 2026, effective 1 January 2027 (secondary); Wikipedia (February 2026) still shows them in committee; no primary text was reached.
• Colorado: the X.AI v. Weiser filing, the US intervention and the 27 April 2026 stay are docket-confirmed. Open: whether the stay reaches SB26-189 (it covers "legislation replacing or amending SB24-205 enacted during this legislative session"), and any docket activity after 27 April. The Attorney General's ADMT and chatbot rules are still in draft.
• Minnesota: in SpaceXAI LLC v. Ellison (8th Cir. No. 26-2806) the injunction pending appeal is undecided; the district-court ruling was not read.
• Federal orders: none of EO 14365's Commerce evaluation, BEAD notice, FCC proceeding or FTC statement was found in the Federal Register, and no DOJ suit against a state AI law was confirmed in any source read. No EO 14409 implementation outputs were found; EO 14434's SI definition proposal is due about 28 November 2026.
• Texas and California dates: TRAIGA's 1 January 2026 start was not found in the enrolled text; SB 53's signing date is 25 or 29 September 2025.
• AI Omnibus formalities: the Official Journal date (24 July 2026) and the Council's approval date rest on law firms; EUR-Lex did not load.
• GPAI Code signatories: the 22 names come from a summarising fetch; re-check them on the Commission page.
• InvestAI (EUR 200 billion) and gigafactory applicants: InvestAI is unverified, and the applicants are not public.
• Anthropic pages: the regional-compliance page could not be re-read on 12 October 2026 (JavaScript-rendered); June 2026 per-capita values for Canada, the UK and EU countries were not extracted, so those tables keep September 2025 values.
• UK: unverified are the formal date and cause of DSIT's replacement, the Growth Zone lead department, the 6 GW target, the GBP 28 billion Growth Zone figure (the Parliament statement returned 403 on 12 October) and ONS's June 2026 rate (29%, single-source news).
• Canada: whether a successor to AIDA is planned (none was found); the Cohere funding round (talks only, versus an unchecked claim of a completed US$1.77 billion Series E); the Cohere data-centre opening; and the close of the Cohere-Aleph Alpha combination.
• Brazil and Mexico: Brazil's post-election vote timing rests on one MobileTime report (not re-read; the Chamber record sets no date); no official revised PBIA total; the two-supercomputer plan and Mexico's law status are single-source; Mexico's plan text was not read.
• Chile and Argentina: Chile's 2026 Senate status is unknown; Argentina's policy was not researched. (Colombia's CONPES figures are now read from the approved PDF, with digits reconstructed from a split text layer.)
• US compute and figures: Abilene's operating capacity (0.6 or 0.3 GW), 2026 capex guidance ($700 or $725 billion), Amazon's XBRL tag, Nvidia's Q2 FY2027 release (figures secondary) and the Pew attitude splits.
• Mistral licensing: the FAQ conflicts with the Apache 2.0 tags on model cards.
• Not researched: national strategies, regulators and compute for the Nordics, the Netherlands, Spain, Poland, Estonia, Ireland and Switzerland (Germany's strategy site lists only the 2020 update and the 2023 Aktionsplan KI); Pew's country values for six Latin American countries; Latinobarómetro; and measured government AI use beyond the UK's self-reported Copilot trials (the US federal inventory and UK "Humphrey" results were not found).

Sources

1. Microsoft: AI Diffusion Report 2026 Q2 (PDF)
2. Eurostat API: isoc_eb_ai, enterprises using AI, 2025
3. IMF DataMapper API: AI Preparedness Index (AI_PI)
4. TOP500: June 2026 list
5. Wikipedia: 2025 in artificial intelligence
6. Federal Register: EO 14363, Launching the Genesis Mission
7. MoFo: Executive order establishes Genesis Mission
8. Federal Register: EO 14365, Ensuring a National Policy Framework for Artificial Intelligence
9. Paul Hastings: Executive order challenging state AI laws
10. Introl: AI preemption executive order and state laws
11. Weintraub: State AI laws under attack
12. Daily Signal: David Sacks out as AI and crypto czar
13. Wikipedia: Sriram Krishnan
14. Wikipedia: Regulation of artificial intelligence
15. NIST: AI RMF Generative AI Profile
16. NIST CAISSI (formerly CAISI)
17. Anthropic: Fable 5 and Mythos 5 access (12 Jun 2026)
18. Al Jazeera: US lifts restrictions on Fable and Mythos models
19. Washington Post: US government will vet users of OpenAI's latest model
20. BBC: Pentagon ceases use of Anthropic products
21. CBS News: xAI Pentagon contract
22. Anthropic: Genesis Mission commitment (8 Oct 2026)
23. Wikipedia: NYT v. Microsoft and OpenAI
24. Bloomberg Law: Anthropic authors' $1.5 billion deal receives final approval
25. Nelson Mullins: Texas HB 149 (TRAIGA)
26. Wikipedia: TRAIGA
27. Clifford Chance: California SB 53
28. NatLawReview: California SB 53, obligations in effect now and into 2027
29. CPPA: ADMT, cybersecurity audit and risk-assessment regulations, approved text (PDF)
30. Colorado General Assembly: SB26-189, Automated Decision-Making Technology
31. Colorado Attorney General: ADMT and Chatbot Safety rulemaking
32. Mondaq/DGL: Colorado's repeal and replacement of its AI Act
33. Wikipedia: RAISE Act
34. Greenberg Traurig: Colorado delays comprehensive AI law
35. Wikipedia: Grok sexual deepfake scandal
36. Wikipedia: 2026 in artificial intelligence
37. NYC Department of Consumer and Worker Protection: Automated employment decision tools (Local Law 144)
38. US Census Bureau: AI use by businesses (May 2026)
39. Federal Reserve FEDS Note: Monitoring AI adoption in the US economy
40. Keyhole Software: US business AI adoption estimate
41. Stanford HAI: AI Index 2026, Economy
42. Pew Research: About half of US adults now use AI chatbots
43. Pew Research: About a quarter of US adults use chatbots daily
44. Gallup: Artificial intelligence indicator
45. TheNextWeb: Pew research on Americans, chatbots and AI regulation
46. Stanford HAI: AI Index 2026, Public Opinion
47. IMF Staff Discussion Note: Gen-AI and the Future of Work
48. Stanford Digital Economy Lab: Canaries in the Coal Mine
49. SEC XBRL: Microsoft capex (PaymentsToAcquirePropertyPlantAndEquipment)
50. SEC XBRL: Alphabet capex (PaymentsToAcquirePropertyPlantAndEquipment)
51. SEC XBRL: Meta capex (PaymentsToAcquirePropertyPlantAndEquipment)
52. SEC XBRL: Amazon capex (PaymentsToAcquireProductiveAssets)
53. Yahoo Finance: Hyperscalers hit $700 billion in 2026
54. Microsoft: FY26 Q3 earnings release
55. NVIDIA: Q2 fiscal 2027 results (figures secondary; release not re-read)
56. Epoch AI: Data on AI data centers
57. Wikipedia: Stargate LLC
58. Bloomberg Law: OpenAI meets key AI computing capacity goal ahead of schedule
59. Epoch AI: OpenAI Stargate, where the US sites stand
60. Wikipedia: OpenAI
61. Stanford HAI: AI Index Report 2026 (PDF)
62. Statistics Canada: Business conditions in Canada, Q2 2026
63. Statistics Canada: AI use release (summary)
64. Anthropic Economic Index: September 2025 report
65. The Logic: Cohere to build AI data centre in Canada
66. Forkast (citing the Globe and Mail): Cohere funding talks
67. Cohere: Cohere and Aleph Alpha sign agreement
68. The Guardian: British Columbia sues OpenAI and Sam Altman
69. Parliament of Canada, LEGISinfo: Bill C-27 (44-1), including the Artificial Intelligence and Data Act
70. ISED: Overview of Canada's National AI Strategy
71. Wikipedia: Evan Solomon
72. Senado Federal: PL 2338/2023 record (Portuguese)
73. Câmara dos Deputados: PL 2338/2023 record (Portuguese)
74. CASRAI: Brazil AI bill status
75. Teletime: AI bill pushed to February 2026 (Portuguese)
76. MobileTime: AI framework vote at year-end (Portuguese)
77. Times Brasil: AI plan foresees R$23 billion (Portuguese)
78. Paraíba Business: Federal government expands AI plan with two supercomputers (Portuguese)
79. IDC Online: How Mexico's new AI plan will work (Spanish)
80. Ecosistema Startup: Mexico AI law 2026 (Spanish)
81. IAPP: Chile's AI bill approved in first constitutional stage (Spanish)
82. DNP: CONPES 4144, Política Nacional de Inteligencia Artificial (PDF)
83. European Commission: AI Act regulatory framework
84. European Commission: Commission starts enforcing AI Act rules and new transparency requirements on 2 August (31 Jul 2026)
85. European Commission: AI literacy, questions and answers
86. Addleshaw Goddard: EU AI Act, AI Omnibus formally adopted
87. Sidley Data Matters: Provisional agreement to delay key AI Act obligations
88. NatLawReview: EU AI Omnibus ("Digital Omnibus on AI") enters force
89. Mobile World Live: EU delays AI rules
90. CASRAI: AI Act deadline delay
91. artificialintelligenceact.eu: Article 50
92. gdpr-info.eu: GDPR Article 22
93. Anadolu Agency: EU opens formal probe into Grok and X recommender systems
94. European Commission: GPAI Code of Practice and signatories
95. EuroHPC JU: AI Gigafactories call launched (30 Jul 2026)
96. StorageNewsletter: EU launches AI gigafactories call
97. WAM: 20% of EU enterprises use AI (citing Eurostat)
98. Implicator.ai: Denmark leads Europe in business AI use at 42%
99. Mistral AI: News
100. Mistral AI: Introducing Mistral Large 4
101. Mistral AI: Pricing and FAQ
102. Sifted: Yann LeCun's AMI Labs funding round
103. n8n: Press page
104. gov.uk: Department for Science, Innovation and Technology (primary; re-read 12 Oct 2026)
105. gov.uk: Ministerial appointments, July 2026 (primary; re-read 12 Oct 2026)
106. AI Security Institute: About
107. Wikipedia: AI Security Institute
108. UK Parliament: AI Growth Zones statement (May 2026; 403 on 12 Oct 2026)
109. gov.uk: AI Growth Zones collection
110. CASRAI: AI Growth Zones (6 GW target unverified)
111. Hacker News search listing: Stargate OpenAI 2026 (headlines)
112. gov.uk: Data (Use and Access) Act 2025, data protection changes
113. Wikipedia: Stability AI
114. Anthropic: Barclays scales Claude
115. Google Cloud: Toolstation product search
116. SiliconANGLE: France announces EUR 109B AI investment
117. The Media Leader: France AI investment commitments (French)
118. LCP: Mistral AI's Arthur Mensch at the National Assembly (French)
119. BMFTR: Federal research ministry home page (German)
120. Comland: Global AI Index 2025 (secondary)
121. Gazzetta Ufficiale: Law No. 132 of 23 September 2025 (Italian)
122. TOP500: November 2025 list
123. Anthropic: Commercial Terms
124. Claude: Regional compliance (read 11 Oct 2026; not re-readable 12 Oct)
125. Anthropic: Claude Frontier Academy
126. Microsoft: AI Diffusion Report 2026 Q1 (PDF)
127. CASRAI: Federal AI moratorium and preemption fight (secondary)
128. Federal Register: EO 14409
129. Federal Register: EO 14434
130. GSA: xAI OneGov agreement (25 Sep 2025)
131. GSA: OpenAI OneGov phase (10 Sep 2026)
132. Texas Legislature: HB 149 (TRAIGA), enrolled text
133. CourtListener: X.AI LLC v. Weiser (docket)
134. CourtListener: SpaceXAI LLC v. Ellison (docket)
135. SecurePrivacy: New York AI law 2026 guide (secondary)
136. Anthropic Economic Index data (June 2026 release)
137. AI Act Service Desk: Implementation timeline
138. artificialintelligenceact.eu: Article 99 (secondary)
139. ONS: Business insights, 8 Jan 2026
140. gov.uk: Andy Burnham, Prime Minister
141. gov.uk: AI investment release (20 Nov 2025)
142. TechRadar: Stargate UK halted (secondary)
143. GOV.UK: AI Opportunities Action Plan, One Year On
144. legislation.gov.uk: DUAA 2025, s.80
145. GOV.UK: Copilot cross-government findings
146. KI-Strategie Deutschland (German)
147. Pew: Global AI attitudes (17 Sep 2026)
148. European Commission: AI at work survey (13 Feb 2025; secondary)
149. Anthropic: Data residency
150. OpenAI: Your data