An AI exercises command over a major power's forces

Where things stand (13 October 2026)

None of the six ministries has published anything that comes close to qualifying since the question opened on 16 September 2026, or earlier. The US is moving fastest on military AI, but every official statement still describes AI as a tool that serves human commanders:

• US Department of War, "Agent Network" release (25 June 2026). It says the system "does not autonomously select or strike targets; it ensures commanders remain in charge of every decision." That is decision support, which the question excludes. https://www.war.gov/News/Releases/Release/Article/4526862/
• Autonomous Warfare Command (announced 30 September 2026; Project Agincourt memo; six-initiatives release). This is a human-led four-star combatant command meant to scale up drones and autonomy, with a target start date of 1 October 2027. Its job is to "integrate Service-provided forces in support of geographic combatant commanders." It does not make an AI the commander. The same applies to the Southern Command autonomous warfare command set up in April 2026. https://media.defense.gov/2026/Sep/30/2004009304/-1/-1/1/PROJECT-AGINCOURT-PATHWAY-TO-THE-AUTONOMOUS-WARFARE-COMMAND.PDF ; https://www.war.gov/News/Releases/Release/Article/4615424/ ; https://www.globalsecurity.org/military/agency/dod/autowar.htm
• NSPM-11 (5 June 2026). It orders faster AI adoption and a rewrite of the Pentagon's autonomous-weapons rule (Directive 3000.09). It also stresses an "unbroken chain of command" and says commanders remain accountable. As of 26 September the rewrite was overdue and unpublished. Even when it appears, a policy document only counts if it also confirms an actual deployment. https://www.whitehouse.gov/presidential-actions/2026/06/national-security-presidential-memorandum-nspm-11/ ; https://www.yahoo.com/news/politics/articles/us-russia-strip-safeguards-un-142645875.html
• Congressional Research Service, IN12669 (updated 5 October 2026). The Pentagon is not publicly known to use frontier AI in autonomous weapons. Senator Kelly's amendment to the Senate defence bill would write "ultimate human responsibility" for lethal force into law. https://www.congress.gov/crs_external_products/IN/PDF/IN12669/IN12669.4.pdf ; https://smallwarsjournal.com/2026/06/17/kill-chain-autonomous-ai-senate/
• China. The People's Liberation Army's official newspaper repeatedly insists that human commanders keep final authority (War on the Rocks, 20 August 2026). At the UN in September 2026, China's foreign ministry said "weapons systems must remain under human control." https://warontherocks.com/chinas-military-says-ai-cant-replace-commanders-xi-is-testing-that/ ; https://dig.watch/updates/south-africa-international-rules-military-ai
• Russia. Its defence ministry talks about AI "elements" in reconnaissance, strike and logistics systems, not AI command. The UK, France and India all keep human-control positions.

Reference class and base rate

No ministry has ever announced AI command of forces in a real operation. The nearest precedents are acknowledged autonomy in single weapons (air defence, loitering munitions) and AI targeting support, such as the Maven system used in the Iran war, where humans approve strikes. Autonomy concepts usually take 5–15 years to go from demonstration to officially acknowledged operational use, and AI command is still at the wargame stage. So the historical base rate is about zero, and only the pace of AI progress raises it.

Paths to YES

1. US (most likely). Under the current leadership the Pentagon's line is "speed wins." It has replaced the "appropriate human judgment" wording with an "any lawful use" standard in AI contracts, and in Geneva it helped remove human-review language from the draft UN text on autonomous weapons (with Russia). The Autonomous Warfare Command should be running in 2027–28, and wars are under way (Iran; strikes on cartels). In a fast-progress world, the Pentagon might boast that an AI commanded an uncrewed task force in a named operation with humans "on the loop."
2. Russia. It could make a propaganda claim about its operation in Ukraine.
3. China. Only in a war, and that would cut against its stated position.

Paths to NO

• Militaries and legal systems treat command as something only humans do.
• Ministries describe AI in hedged terms such as "commanders remain in charge."
• Details of real operations are classified.
• International human-control diplomacy is gathering pace (US–China AI dialogue; UN autonomous-weapons talks in November 2026).
• Congressional pressure, and possibly a more cautious US administration after 2028.
• The question requires explicit wording on four elements (delegated authority over units, the named operation or theatre, the forces involved, and no prior human approval), and it excludes drone-swarm coordination and generic "AI command and control" claims. Most plausible announcements would therefore fail.

Rough numbers and hazard shape

A scenario mix gives about 6%:

Scenario	Weight	Chance of a qualifying announcement by 2031
Fast AI progress	~30%	~15%
Moderate progress	~50%	~3%
Slow progress or clampdown	~20%	~0.5%

I turned this into quarterly hazards: about 0.08% for the rest of Q4 2026 (nothing in the pipeline; the Project Meridian report and the Directive 3000.09 rewrite cannot confirm a deployment). The hazard then rises steadily to about 0.48% per quarter in 2031 as capability and autonomous-force structures mature. Cumulative probability is about 0.6% by end-2027, 1.4% by end-2028, 2.5% by end-2029, 4.0% by end-2030 and 5.7% by end-2031. These figures come from the cumulative product of (1 − hazard).

Strongest case against my estimate

AI progress in 2026 is extraordinary. The US government officially calls AI "Super Intelligence" and is proposing an "AI Force," and its defence leadership openly plays down safeguards. Five years is long enough for AI-run autonomous formations to become routine and for a boastful minister to describe them plainly. That is why the hazard rises over time rather than staying near zero. Even so, every 2026 statement, including the most aggressive ones, still puts humans in command.

Question. By the last day of each quarter, has the defence ministry of the US, China, Russia, the UK, France, or India announced that an AI system exercises command over its forces in a theatre or a named operation, with human commanders in a supervisory role only?

Definitions.

• AI system: a system described by the announcing ministry as using artificial intelligence to make command decisions; no particular developer, model architecture, or level of general intelligence is required.
• Exercises command: has delegated authority to decide and issue operational orders to subordinate military units or formations, including assigning or changing missions and directing their execution. Its orders take effect without prior human approval of each order. Merely preparing plans, recommending targets, relaying human orders, or controlling the technical execution of a human-selected engagement is not command for this question.
• In a theatre or a named operation: the system holds a command role over forces assigned to an identified geographical military theatre or an officially named real-world military operation. It need not command every national force in that theatre, but must command an identifiable force or formation, rather than merely operate an individual weapon or coordinate platforms executing a fixed human-assigned task. Forces may be crewed, uncrewed, or mixed; no minimum troop count, duration, or requirement to engage in combat applies.
• Human commanders in a supervisory role only: within the announced command remit, humans monitor, audit, veto, override, or terminate the system's decisions rather than routinely select or approve operational orders before execution. Humans may set the overall mission, rules of engagement, geographical limits, and other standing constraints, and retain legal responsibility. A required affirmative human sign-off on operational orders does not qualify, even if described as a formality. Human command outside the AI's delegated remit does not disqualify it.
• Announce: an institutional statement that the arrangement is in effect or was actually used, not merely that it is technically possible, authorised for future use, planned, or being procured.

Resolves YES if a qualifying ministry publication explicitly establishes all of the above: an AI system actually exercising command, the theatre or named operation and forces under its command, and the absence of routine prior human approval within its command remit. These facts may be established across multiple official ministry publications referring to the same deployment; resolution occurs when the last necessary confirmation is published. An official doctrine document counts only if it also confirms an actual qualifying deployment. The announcement itself is the event being forecast; independent verification of battlefield performance is not required.

Resolves NO / Exclusions.

• No qualifying announcement by the quarter-end.
• Simulations, wargames, training exercises, laboratory experiments, demonstrations, and trials without a real operational command assignment, even if they have an operation-style name or involve live equipment.
• Decision-support or campaign-planning systems where humans still decide which orders to issue; generic claims of “AI command and control,” “machine-speed command,” or “autonomous mission command” without the specified authority and human role.
• Autonomous targeting, firing, navigation, drone-swarm coordination, or cyber activity alone, without the defined command over forces.
• Contracts, future doctrine, research proposals, vendor claims, media reports, and statements by other governments about the country's forces.

Resolving source. Official ministry-issued releases, reports, doctrine documents, or ministry-published transcripts of statements by the minister or an authorised spokesperson, from: the US Department of Defense/Department of War (defense.gov or war.gov); China's Ministry of National Defense (mod.gov.cn or eng.mod.gov.cn); Russia's Ministry of Defence (mil.ru or eng.mil.ru); the UK Ministry of Defence (gov.uk/government/organisations/ministry-of-defence and its publications); France's Ministère des Armées et des Anciens combattants (defense.gouv.fr); or India's Ministry of Defence (mod.gov.in, including Ministry of Defence releases issued through the Press Information Bureau at pib.gov.in). Renamed successor ministries and official successor domains count. A service, command, contractor, or military-academic publication counts only if the ministry expressly adopts its claim as an institutional announcement, not merely hosts or links to it. This is a narrower source requirement than Public confirmation (A2); third-party confirmation cannot substitute. Apply Resolution date (A3) and the Resolution window.