A G20 government says AI makes most of its decisions

Where things stand (11 Oct 2026)

I found no qualifying statement from any G20 head of government or legislature, before or after the 16 Sep 2026 opening date. Nothing official comes close. Even the most aggressive governments describe AI's role as a share of services or operations, or as decision support. None claims a share of government-wide decisions:

• UAE (not a G20 member). This is the boldest official framing anywhere. On 23 Apr 2026 Prime Minister Mohammed bin Rashid set a target to move "50% of UAE Government sectors and services within two years to Agentic AI for autonomous execution and decision-making" (Dubai Media Office; Fox News, 3 May 2026). Even if a G20 government copied it and later said it had hit the target, that would fail here. It is a share of sectors and services (adoption), not of decisions, and 50% does not exceed 50%.
• South Korea. The "AI-Driven Democratic Government" strategy (Cabinet, 25 Aug 2026) talks about AI consultations, drafting and "collaboration between public officials and AI" (Asia Business Daily).
• Russia. The updated public-administration digital strategy (July 2026) and the 2027–2030 AI targets aim AI at routine tasks and document handling (ComNews, 16 Jul 2026; government.ru).
• Argentina. Milei's AI "social digital twin" project (Gemelo Digital Social) is pitched as optimising and predicting decisions (El País, 23 May 2026).
• United States. America.gov (launched 29 Sep 2026) is an AI front door for information (FedScoop). Executive Order 14434 renames AI "Super Intelligence" across the executive branch (whitehouse.gov). The order defines "Super Intelligence" as AI, so I read a US claim using that term as naming an AI system. Its framing is still that AI "amplifies human ingenuity."
• OECD. Governments are adding human-approval safeguards for AI agents, and fewer than 4 in 10 people trust government AI use (OECD.AI, Sep 2026).

Base rate

No head of government has ever officially claimed that computers, algorithms or AI make most of a national government's decisions. The closest case is Albania's "AI minister" Diella. It covered a single area (procurement), Albania is not in the G20, and a year later the system still was not running (BIRN, 14 Aug 2026). The base rate is effectively zero.

Paths to YES

1. Fast AI progress, then deep government adoption by 2029–31, plus a leader who boasts about it. This world does look fast. OpenAI's chief scientist expects recursive self-improvement (An Alien Mind), and Amodei says progress sped up sharply (Guardian, 12 Sep 2026). The question counts decisions by number, with no weighting by importance. So once AI handles routine casework (tax, benefits, permits), the claim could be literally true. Likely speakers would be a populist or tech-booster leader: the US president, Milei if re-elected, or Saudi Arabia's prime minister, Mohammed bin Salman.
2. Off-the-cuff exaggeration published in an official transcript. For example, a US president saying "SI makes most of the decisions now."
3. A legislature formally adopting such a finding. This is very remote.

Paths to NO, and the strongest counterpoint

Leaders have strong reasons to keep humans formally in charge: legal accountability and due process, data-protection rules on automated decisions in the EU, UK, Brazil, Korea and China, and low public trust. The 2026 incidents with rogue AI agents add to this; OpenAI has paused training twice (Guardian, 27 Sep 2026). The question's exclusions are tight: claims about routine decisions only, AI-assisted decisions, adoption shares, targets and jokes all fail. Only the head of government counts, so Putin and Xi do not (Russia's and China's heads of government are their premiers), and only official government or legislature sources count.

The strongest case for a higher number is the fast-takeoff world combined with 19 governments and several changes of leader over five years. I put roughly 15% on AI actually making most decisions in some G20 government by 2031. Given that, I put about 10–15% on a qualifying statement being made and published, plus about 0.5–1% for exaggeration and other paths. That gives roughly 3% in total.

G20 membership changes (the US barring South Africa, Poland's bid) don't move this; I found no sign of the UAE seeking membership.

Computation

I set quarterly hazards, the chance of a first qualifying publication in each quarter if none has appeared yet, and accumulated them in Python as 1 − ∏(1 − h). They run from 0.05% for Q4 2026 to 0.08% by Q4 2027, 0.12% by Q4 2028, 0.16% by Q4 2029, 0.21% by Q4 2030 and 0.25% by Q4 2031. They rise because adoption trails capability. The cumulative probability is about 0.3% by end-2027, 0.7% by end-2028, 1.3% by end-2029, 2.1% by end-2030 and 3.0% by end-2031.

Question. By the last day of each quarter, has the head of government or the legislature of any G20 country stated officially that most decisions of its national government are made by AI systems?

Definitions.

• G20 country: a sovereign country that is a G20 member on the publication date. The opening-date countries are Argentina, Australia, Brazil, Canada, China, France, Germany, India, Indonesia, Italy, Japan, the Republic of Korea, Mexico, Russia, Saudi Arabia, South Africa, Türkiye, the United Kingdom, and the United States (see the official G20 membership list). EU or African Union membership alone does not qualify, and neither union is a country.
• Head of government: the incumbent national prime minister, premier, or other officeholder constitutionally heading the national executive, including a president where that office also heads the government. A head of state without that role does not qualify.
• Legislature: the national legislative body acting institutionally through a formally adopted resolution, law, declaration, or report. Adoption by the plenary of either chamber of a bicameral legislature qualifies; statements by individual legislators, presiding officers, parties, or committees alone do not.
• Most decisions of its national government: more than half of the decisions of the central executive government as a whole, including its ministries and agencies. An explicit government-wide statement using “most,” “a majority,” or an unambiguously stronger formulation qualifies without a numerical tally. A numerical claim must exceed 50% and refer to government-wide decisions, not a share of agencies, staff, spending, or AI adoption. No separate weighting by importance is required.
• Made by AI systems: AI is stated to select the substantive outcomes of decisions, rather than merely provide information, draft documents, or advise human decision-makers. Human formal sign-off, responsibility, or override powers do not disqualify a statement that expressly attributes the actual decisions to AI. The statement must identify the decision-maker as artificial intelligence or an AI system; references solely to software, digitisation, or automation do not suffice.
• Official statement: an affirmative factual assertion made by the head of government in their official capacity and published in an official speech transcript, recording, signed statement, or expressly attributed passage of an official report; or an assertion adopted by the legislature as defined above. An official publication quoting someone else's claim without adopting it does not qualify.

Resolves YES if a qualifying official statement says that AI systems currently make most decisions of that country's national government, or did so during a specified period ending after the opening date. This is a question about an official declaration: independent verification of the underlying decision share is not required. The statement must concern actual government operations, not a hypothetical exercise.

Exclusions (resolve NO).

• Plans, promises, authorisations, targets, forecasts, questions, jokes, or conditional assertions about future AI government.
• Claims confined to a particular ministry, programme, decision category, pilot, or subnational government, even if AI makes all decisions there.
• Claims that most decisions are AI-assisted, informed by AI, or reviewed by AI, without attributing the substantive decisions themselves to AI.
• Claims merely that AI makes most routine decisions, or most decisions suitable for automation, without asserting a majority of national-government decisions overall.
• Statements by ministers other than the head of government, civil servants, political candidates, former officeholders, or external observers, unless expressly adopted by a qualifying speaker or legislature.
• No qualifying publication by the relevant quarter-end.

Resolving source. Only the country's official head-of-government or executive website, national government gazette, or national legislature's official website, adopted documents, or parliamentary record may establish the statement. This is a narrower source rule than Public confirmation (A2); third-party reporting alone does not qualify. Apply the Resolution window and Resolution date (A3) conventions. Apply the Observation period convention to numerical statistics; a publication only repeating figures for a period ending on or before the opening date does not qualify.