takeoff.watch

An AI writes the plan for a military campaign that is carried out

Will a government state officially that an AI system produced the operational plan for a military campaign that its forces then carried out?

Chance by Q4 2031

17%

Cumulative probability

0%25%50%75%100%
  • Latest forecast

Forecast history

Forecast made on 14 Oct 2026

Latest of 5 forecasts · Methodology v2.1

Model reasoning

Forecast made 14 Oct 2026 · Methodology v2.1

Militaries now openly credit artificial intelligence with speeding up targeting and planning. But every official account still says the AI assists and humans plan and decide, and no government has yet said an AI produced a campaign plan that its forces then carried out. I put the chance at under 1% this year, rising to about 17% by the end of 2031. What would most change this: a US official boasting that a renewed Iran campaign ran on an AI-written plan.

Where things stand (as of 14 Oct 2026)

I found no qualifying statement since the window opened on 16 Sep 2026. Militaries now talk about AI in war more openly than ever, but every official account I found describes AI as helping with targeting or planning. The criteria exclude that.

  • United States. CENTCOM's Adm. Cooper credited "AI war planning" in Operation Epic Fury, but added that "humans will always make final decisions" (CENTCOM video, 11 Mar 2026). The Pentagon's AI chief, Cameron Stanley, said in a sworn statement that Maven "frontier workflows enabled" 2,000 munitions against 2,000 targets in 96 hours (June 2026). Both are about targeting. Two combatant commanders (Transportation and Strategic Command) said AI "generate[s] options… faster" for the commander (June 2026). That counts only as advice on options.
  • Israel. IDF officials say AI tools "help build effective attack plans," but the IDF "does not allow AI systems to make autonomous decisions" (Ctech, Mar 2026). Its response to the film NAZA (Oct 2026) says the technology "assist[s]" analysts.
  • Ukraine. The defence ministry's SPECTR platform produces "recommendations for operational planning" and is still being tested. Fedorov, the minister most inclined to tech publicity, was dismissed in July 2026.
  • UK. The ASGARD system cut a corps planning cycle from 72 hours to 1, but that was in an exercise, which is excluded.
  • China. In the PLA Daily, articles asserting the human commander's final authority over AI rose from 45 in 2018 to 153 in 2025 (War on the Rocks, Aug 2026).

Reference class and base rate

From 2021 to 2026, about 20 quarters, governments made many official AI-in-war claims: Israel's "first AI war" and "target factory", India's claim of 94% accuracy in Operation Sindoor, and the US claims about Epic Fury. None said an AI system produced a campaign's operational plan. A Jeffreys-style estimate from zero events in about 20 quarters gives roughly 2.5% per quarter at most. Capability was lower in the past, but the norm against such claims has hardened rather than weakened. The closest analogy is targeting. Israel officially said its Gospel system "produces targets" about two to four years after AI targeting became practical. Saying AI produced a campaign plan is a bigger step, because it moves credit and accountability away from commanders.

Paths to YES

  1. United States. The Pentagon's AI strategy includes an "Agent Network" project for AI agents "from campaign planning to kill chain execution" (Jan 2026). The administration has renamed AI "super intelligence", and Hegseth speaks of "SI-enabled targeting." An Autonomous Warfare Command is due by Oct 2027. A short air campaign against Iran has been drafted, and Trump has ruled out strikes only until the 3 Nov midterms. A boast that "SI generated the campaign plan; the commander approved it" would qualify, because human selection and approval are allowed.
  2. Israel, Ukraine and India all have a record of public AI claims and conflicts that are ongoing or likely.
  3. Propaganda or head-of-state exaggeration (Trump, Russia, Iran) recorded in an official transcript.
  4. Routine after-action reports or testimony after 2029, once AI planning is normal.

Paths to NO and the main counterweights

  • Officials consistently say "the commander decides."
  • Accountability pressure is high: the Minab school strike, where investigators reportedly found over-reliance on AI; the near-interception of a Chinese ship after an AI-generated false report; and Senate requests for inspector-general probes (Sept 2026).
  • The criteria are strict. The plan must be campaign-level. Single strikes or raids, target lists and air tasking orders do not count, and neither does "AI-enabled" or "AI-assisted." The source must be official.
  • A US administration after January 2029 may be more cautious.

Strongest case against my estimate: AI capability is moving very fast. There are reports of autonomous agents "breaking containment," and the US government now officially calls AI "super intelligence." An AI-maximalist US leadership in an active war could make an explicit boast well before 2028. I hold the hazard at about 1% per quarter because every official account so far stops at "assist," but I have raised the later quarters for capability growth.

Interpretation and model

I assume a resolver would treat the US term "super intelligence" as identifying an AI system. I treat a statement that AI generated campaign courses of action, a commander selected one, and the campaign was executed as qualifying.

The quarterly hazards are 0.5% for the rest of 2026, 0.6–0.8% in 2027, 0.8–0.9% in 2028–29, and 0.9–1.0% in 2030–31. Accumulated as 1 − ∏(1 − h), these give about 3% by end-2027, 7% by end-2028, 10% by end-2029 and about 17% by end-2031. I computed them in Python.

Sources

Resolution criteria

Opened 16 Sep 2026Not yet resolved

Question. By the last day of each quarter, has a government stated officially that an AI system produced the operational plan for a military campaign that its forces then carried out?

Definitions.

  • Government: the national government of any state, including a government with limited international recognition; statements by non-state armed groups or subnational governments do not count.
  • AI system: a computational system identified as artificial intelligence or machine learning in the qualifying official account. It need not be autonomous or developed by a covered developer.
  • Military campaign: a coordinated series of military operations directed toward a common strategic or operational objective against an adversary. A named operation counts if it comprises such a series; a single strike, raid, engagement, or targeting cycle does not. Campaigns conducted in any military domain, including cyberspace, count, as do a state's contributions to coalition campaigns.
  • Operational plan: the campaign-level scheme specifying how forces and operations are to be employed and coordinated to achieve the campaign's objectives, rather than just intelligence assessments, target lists, individual mission plans, logistics schedules, or advice on options.
  • Produced: the official account explicitly attributes generation of the campaign's operational plan to the AI system. Human provision of objectives and constraints, selection among AI-generated plans, review, approval, and subsequent adjustments are permitted. Merely assisting human planners, drafting prose from a human-designed plan, or generating an unadopted option is insufficient. Where humans revised an AI-generated plan, the account must still identify the executed campaign plan as the AI-generated plan, not merely as informed by AI output.
  • Then carried out: the account confirms that the state's armed forces actually executed a series of operations under that plan after its generation. The campaign need not have ended or succeeded, and not every planned operation must have occurred. Approval, orders to deploy, or execution of only an initial isolated strike is insufficient.

Resolves YES if one or more qualifying official publications, taken together, explicitly establish all of the following for the same identifiable campaign: an AI system generated its operational plan; the government adopted that plan; and its forces subsequently carried out the campaign under it. The plan itself and the system's technical identity need not be disclosed. This question concerns an official attribution; independent verification of the government's claim is not required.

Resolves NO / Exclusions. Resolve NO for a horizon if no qualifying statement has been published by that horizon. Exclude exercises, simulations, wargames, demonstrations, hypothetical plans, procurement announcements, future intentions, and claims that AI merely supported campaign planning or targeting. General descriptions such as “AI-powered war” do not establish authorship of the operational plan. Attribution by a vendor, another government, researchers, or journalists alone does not count.

Resolving source. Official publications or officially published transcripts or recordings from the government's head of state or head of government, cabinet office, defence ministry (including an equivalently named war ministry), or armed forces, including their operational commands and authorized spokespeople. Official reports and testimony published by those bodies qualify; personal posts, leaked material, anonymous briefings, and press quotations without an official published record do not. These are the narrower resolving sources for this question rather than the broader Public confirmation convention A2. Apply the Resolution window and Resolution date convention A3. If several publications are needed, use the publication date of the last one needed to establish all conditions. A post-opening official retrospective account may qualify even if the campaign occurred earlier.

Download all forecasts (JSON)