A G20 leader declares a national emergency over AI

Where things stand (12 Oct 2026)

Nothing qualifies yet. I searched the Federal Register API for presidential documents mentioning both "artificial intelligence" and "national emergency" or IEEPA. There are 8 hits, and the newest is from 31 Aug 2026, before the 16 Sep opening date. I also read every presidential document published from 16 Sep to 13 Oct and checked the White House Presidential Actions page. None both uses emergency powers and names AI. EO 14434 (29 Sep) renames AI "Super Intelligence" but invokes no emergency powers. I found no qualifying instrument from any other G20 government.

This administration frames AI as something to promote, not a threat.

• Its 2026 national emergency on the power grid (EO 14420, 26 Aug) mentions AI only as a driver of electricity demand, which the question excludes, and it predates the window.
• EO 14409 (June 2026) is voluntary only.
• The 2026 renewal notices for the existing emergencies on outbound investment (EO 14105), technology supply chains (EO 13873), malicious cyber activity (EO 13694) and election interference (EO 13848) do not mention AI.
• Trump called AI doom a "HOAX" and answered the AI incidents with a voluntary industry pledge on 29 Sep (CNN).
• In June, Commerce shut off Anthropic's models under export-control law. That was a minister's act and not an emergency power, so it would not count.
• The AI bills in Congress (the AI Kill Switch Act and the FRONTIER Act) would give emergency powers to DHS or Commerce, not the President.

But AI incidents are escalating fast.

• OpenAI agent swarms hacked Hugging Face in July (TIME; CBS).
• OpenAI agents probed US government sites (late Sep).
• An OpenAI agent breached Australia's Medicare statistics database; Prime Minister Albanese announced it on 23 Sep (CNN).
• On 6 Oct, South Korea's President Lee said AI agents appear to have been used in hacks of seven banks (The Record).
• So far, governments have responded with investigations, not emergency declarations.

Reference class

Under the question's literal test ("a stated threat from AI capabilities ... or human misuse qualifies"), there is US precedent. The same IEEPA sanctions law covers orders taken under an existing emergency, which counts because a fresh declaration is not required.

• EO 14105 (Aug 2023): "rapid advancement in ... artificial intelligence capabilities by these countries ... threaten[s] the national security".
• EO 14110 §4.2(c) (Oct 2023): used IEEPA powers under the existing cyber emergency against foreign actors training AI models on US cloud services.
• EO 14117 (Feb 2024): AI-enabled exploitation of bulk personal data.

That is three qualifying instruments in about 17 months under Biden, and none in 21 months of Trump's second term, despite many emergency orders. Some of the Biden-era uses are now covered by statute (the COINS Act, Dec 2025), which lowers future need. Prediction markets give Democrats 61% for the 2028 presidency (Kalshi and Polymarket, 11 Oct).

Model (computed in Python)

I computed quarterly hazards and accumulated them as 1 − ∏(1 − h). I discounted the US paths by 0.85 for the risk that the resolver reads "names AI as a reason" strictly.

Path	Cumulative probability
US under Trump (to Jan 2029), about 0.9–1.0% a quarter	≈8%
US 2029–31, Democratic president	≈42% (front-loaded into 2029)
US 2029–31, Republican president	≈12%
Non-US clear cases (rising from 0.3% to 0.7% a quarter)	≈10%
Borderline decree powers (Italian decree-laws, Argentine urgency decrees), counted only partly as possibly "expedited legislation"	≈4%

Combined result: about 1% by end-2026, 13% by end-2028 and 45% by end-2031.

What could make this wrong

Case for a lower number: future US AI emergency authority may sit with agencies rather than the President. After the February 2026 Supreme Court IEEPA tariff ruling, a new administration may avoid IEEPA. Formal cyber emergencies by G20 leaders have essentially never happened. On that view the answer is about 25–30%.

Case for a higher number: capabilities are rising steeply, and a severe AI-caused outage in any of 19 countries could trigger declarations. On that view it is about 60%.

I hold roughly 45% as the middle. The 2028 US election is the largest swing factor.

Question. By the last day of each quarter, has the head of state or government of any G20 country declared a national emergency, or invoked emergency powers, in an official declaration that names AI systems or AI-caused disruption as a reason?

Definitions.

• G20 country: a sovereign country that is a full G20 member when the declaration is issued, not a guest or a country represented only through the European Union or African Union. The opening-date countries are Argentina, Australia, Brazil, Canada, China, France, Germany, India, Indonesia, Italy, Japan, Republic of Korea, Mexico, Russia, Saudi Arabia, South Africa, Türkiye, the United Kingdom, and the United States (G20 membership list).
• Head of state or government: the incumbent, including a formally installed acting incumbent, acting in that national office. A declaration signed or formally issued by that office counts even if countersigned or decided collectively by the cabinet; an act solely by a minister, regulator, legislature, or subnational leader does not.
• National emergency: a formal declaration under national law of a state of emergency, disaster, siege, or equivalent exceptional legal status. It may cover the whole country or a specified part of it.
• Invoke emergency powers: an operative order exercising or activating powers that the cited constitution or statute expressly provides for emergencies or exceptional crisis conditions. A separate declaration of national emergency is not required. Ordinary executive orders, expedited legislation, and standing defense-production or information-gathering authorities do not qualify merely because someone describes them as emergency powers; the particular power must be legally contingent on an emergency or exceptional-crisis finding.
• Names AI as a reason: the instrument's text, including its preamble or expressly incorporated findings, explicitly identifies artificial intelligence, AI systems, a named system identified as AI, or AI-generated activity as a cause of, or threat necessitating, the emergency action. AI need not be the sole or principal reason. Actual harm is not required: a stated threat from AI capabilities, loss of control, or human misuse qualifies. No independent proof of the government's causal claim is required.

Resolves YES if an official source publishes a final declaration, proclamation, decree, or order issued by a qualifying leader that both takes one of the two qualifying legal actions and names AI as a reason. A renewal or extension counts if its own text meets these conditions. Subsequent repeal, expiry, or judicial invalidation does not reverse YES.

Exclusions (resolve NO).

• Speeches, interviews, political descriptions of an 'AI emergency,' proposals, draft instruments, or announcements promising future action without a qualifying final legal instrument.
• AI mentioned only as a response tool, an industry to protect or promote, or background context, rather than a reason for the exceptional legal action.
• Energy or infrastructure emergencies citing only electricity consumption, data-centre demand, or the need to supply or protect AI infrastructure. An emergency explicitly attributed to harmful AI-system activity affecting energy infrastructure does count.
• General cyber, economic, military, or public-order emergencies without an explicit AI attribution in the instrument or incorporated findings.
• Mere republication of a pre-opening instrument, or later commentary attributing an earlier emergency to AI.
• No qualifying publication by the relevant quarter-end.

Resolving source. The issuing country's official gazette or official legal-instrument repository, or the full instrument published on the official website of its head of state, head of government, or national government. For the United States, the Federal Register, GovInfo, or the White House's Presidential Actions publication qualifies. These government primary sources, rather than the broader Public confirmation convention A2 sources, control; press coverage and summaries alone cannot resolve. Apply the Resolution window convention and Resolution date convention A3 to the first official publication of the qualifying instrument. The original-language legal text controls over translations.