23 entries on the record in technology & software — 10 pieces of evidence of adoption, 6 regulations, 4 pieces of evidence of impact, 3 deployments. Each one is traced to a named source and dated. Maintained continuously by the Industrial Research Unit; the record was last updated 14 Sep 2026.
State of California — California Governor Gavin Newsom signed two bills his office calls first-in-the-nation standards for independent assessment of AI systems. Senate Bill 813 frames independent verification organisations that check AI systems and models for state-law compliance; Assembly Bill 1405 creates a state registry of AI auditors with independence, transparency and integrity standards.
B3 — B3, Brazil's stock exchange, made Anysphere's Cursor its official AI coding assistant for engineering, widening it to about 600 employees after active developers grew from 34 to 154 between May and July. Luiza Sangalli, B3's superintendent of artificial intelligence, says application delivery time fell by about 35%, and the exchange now accepts roughly 1.5 million lines of AI-assisted code a month. Human review, testing and security checks still apply, and no independent measurement of the 35% figure has been published.
Government of Japan — Japan's government has adopted principles on AI training data disclosure, pressing generative AI operators to share basic facts with users, rights holders and the public. Operators should publish an overview of their automated online data-collection tools, tell rights holders on request whether their works were used, and file their implementation with the Cabinet Office, updating it yearly. The non-binding principles carry no penalties and cover foreign as well as Japanese firms offering AI services in the country.
Infosystems Jet and Smart Ranking — 86% of large Russian companies now use or pilot large language models, close to McKinsey's global rate of 88%, per research unveiled at the State of AI Russia 2026 event by Infosystems Jet and Smart Ranking. This LLM adoption in Russian companies drops off sharply at higher autonomy: semi-autonomous agents run at only 15% of firms, and fully autonomous systems at just 8%. Respondents named GPU expense (58%) and legal exposure (56%) as the leading obstacles to wider enterprise AI adoption.
Réunion24 Aug 2026Pigé survey for Le Quotidien — AI tools Réunion adoption stands at 73% of internet users, a survey by Pigé for Le Quotidien found, matching levels seen in mainland France. The poll of 663 respondents aged 15 and over found artificial intelligence use has overtaken e-commerce in uptake speed, despite arriving on the island only in late 2022. Convenience and performance ratings for AI beat search engines, though trust and reliability lagged by up to 25 points.
Equipos Consultores — Equipos Consultores finds half of Uruguay's adults have never used AI tools; a quarter are daily or fairly frequent users and the rest occasionally. AI use in Uruguay tracks education: almost nine in ten university graduates have tried it, but eight in ten people with only primary schooling never have. The survey of 705 adults on AI habits ran from 23 June to 8 July 2026.
Colorado Department of Law — The Colorado Department of Law published draft rules on 11 August for the state's Automated Decision-Making Technology Act and its conversational AI service provisions, setting compliance duties for developers and deployers. Signed by Governor Jared Polis on 14 May as SB 26-189, the ADMT Act repealed and replaced the 2024 Colorado AI Act with a narrower framework on transparency and disclosure of adverse outcomes from automated decisions, dropping mandatory risk-management programmes, impact assessments and the duty of reasonable care. Comments are invited before a 26 October public hearing, and the Act applies from 1 January 2027 whether or not rulemaking is finished.
State of California — California's AI Transparency Act took effect on 2 August, obliging publicly available generative AI services with over one million monthly users in California to add latent disclosures to AI-generated audio, video and images, offer an optional manifest disclosure and publish a free AI detection tool. Passed as SB 942 in 2024 and amended by AB 853 in 2025, it carries civil penalties of $5,000 per violation, each day counting separately, with no private right of action. Hosting and large online platforms face further duties from 1 January 2027, and capture-device makers from 2028.
Slovenia4 Aug 2026DOBA Business School — Slovenia tops the AI maturity index among Slovenia, Croatia and Serbia, scoring 50.85 out of 100 in DOBA's DAIMI 2026 survey - nearest of the three to the 'integrated' benchmark. Firms were surveyed between March and May 2026 across five dimensions: strategy, governance, workforce skills, deployment and business impact. Even progress on all fronts drives this AI adoption readiness lead, rather than strength in one area.
Moldova Innovation Technology Park (MITP) — A new poll shows 78% of firms at Moldova Innovation Technology Park (MITP) have woven artificial intelligence into daily operations, reflecting broad AI adoption in Moldova IT Park. Researchers surveyed nearly 2,800 businesses between April and May 2026: 37% run it organisation-wide, 41% apply it to select tasks, 10% are trialling it, and 11% reported none. Findings point to workplace AI tools spreading across Moldova's tech sector.
HKIRC — A poll of 800 organisations by Hong Kong Internet Registration Corporation found 94% had adopted AI tools at work, reflecting widespread AI adoption in Hong Kong businesses. Yet 63% had no formal policy covering workplace AI tools and 68% skipped staff training on the technology. Unauthorised AI tools were in use by staff at nearly half the firms surveyed, without company approval.
Afghanistan4 May 20261,154 respondents across 32 provinces went into a national survey of AI use in Afghanistan, finding 88% of young Afghans familiar with artificial intelligence and 66.6% of them regular users. Media outlet Salam Afghanistan and Salam Watandar Radio's eight-month fieldwork, split 50.6% men and 49.4% women, found 84.1% had sought AI advice at least once and 82% now lean less on friends and relatives, with 58.1% uneasy about machines making decisions.
Vietnam24 Feb 2026National Assembly of Vietnam — Vietnam's first standalone AI law took effect on 1 March 2026, enacted on 10 December 2025, grading AI systems high, medium or low risk by effect on safety and human rights, field of application, user scope and scale of effect. Providers must tag AI-generated audio, image and video in machine-readable form, and deployers must flag it where it might blur whether events or people are real. Baker McKenzie says most existing providers and deployers have until 1 March 2027, and the Ministry of Science and Technology will issue implementing guidance.
North Macedonia11 Feb 2026Eurostat — North Macedonia trails the EU in AI tools adoption among young people, Eurostat's 2025 data shows: 45.4% of those aged 16-24 use such tools, against 63.8% across the bloc. Generative AI usage among all residents aged 16-74 reached just 22% that year, leaving a wide gap versus the EU average for both youth and the wider population.
Infocomm Media Development Authority — Singapore launched the Model AI Governance Framework for Agentic AI at the World Economic Forum in Davos, covering AI agents that reason and act autonomously for users. The Infocomm Media Development Authority sets out technical and non-technical measures for agentic AI governance across four dimensions: bounding risks up front, meaningful human accountability through checkpoints, technical controls over the agent's lifecycle and end-user education. Minister for digital development and information Josephine Teo announced it.
The US Federal Trade Commission issued 6(b) study orders, by a 3-0 vote on 11 September 2025, to OpenAI, Character Technologies, Meta, Instagram, Snap, xAI and Alphabet, examining possible harm to children and teenagers from AI 'companion' chatbots. It also checks children's online-privacy compliance, running alongside wrongful-death litigation against OpenAI (Raine v OpenAI, filed August 2025), a defining AI-safety case of 2025-26.
About 44% of AI code-generation tasks added a security vulnerability in Veracode's 2026 GenAI Code Security Report, whose average security 'pass rate', about 56%, is essentially flat on last year's 55%. AI-written code is no more secure even as its volume surges.
AI adoption follows a 'J-curve', productivity first dipping during rollout and then rising, Google Cloud's 2026 DORA research ('The ROI of AI-Assisted Software Development') finds. It confirms AI magnifies whatever strengths and weaknesses an organisation already has.
METR's randomised trial of 16 experienced open-source developers (Cursor Pro, Claude 3.5/3.7; Feb-June 2025) showed tasks taking about 19% longer with those early-2025 AI tools, although participants predicted gains and still felt quicker afterwards, so self-reported AI coding productivity cannot be trusted. METR's follow-up cohort (from Aug 2025) and 24 Feb 2026 update point to probable gains with today's tools, lower bounds of about 18% for returning developers and 4% for new recruits, with developers likely quicker in early 2026 than early 2025. Because developers who value AI most increasingly refuse to work without it, selection effects make task-level developer productivity hard to gauge, and METR is reworking its study design. Another METR survey of 349 technical workers (Feb-Apr 2026) recorded median self-reported gains of 1.4-2x in work value and 3x in speed, which METR warns are self-reports rather than measured effects; its own staff gave the smallest figures.
In 2025, 84% of developers used or planned to use AI tools in development, up from 76% in 2024. 51% of professional developers now use them daily.
AI adoption went with higher software delivery throughput but, again, lower stability in Google Cloud's 2025 DORA report. It surveyed nearly 5,000 technology professionals, 90% of whom use AI at work.
Developer trust in AI is falling as adoption rises. In 2025, 46% of developers actively distrusted AI tools' accuracy against 33% who trusted it, and favourable sentiment fell to 60%, from over 70% in 2023 and 2024.
AI now writes about 75 per cent of Google's new code, which human engineers then review and approve, chief executive Sundar Pichai said at Google Cloud Next in April 2026, up from around 50 per cent the previous autumn. The on-record vendor claim rests on a loose definition of 'AI-generated'.
United States
Brazil
Japan
Russia
Uruguay
Moldova
Hong Kong
Singapore