A major power says it fields fully autonomous lethal weapons

Current status (as of 11 Oct 2026)

I found no qualifying publication since the question opened on 16 Sep 2026. Every covered government's official line still says a human makes lethal targeting decisions:

• US. NSPM-11 (5 Jun 2026) ordered the Secretary of War to update DoDD 3000.09 within 90 days. The update has not appeared yet. On 22 Sep 2026 Under Secretary Colby told the Senate Armed Services Committee that DoW "is ... updating the 3000.09 policy ... as directed by NSPM-11" (statement). The Congressional Research Service still cites the January 2023 version (IN12669, 5 Oct 2026) and says senior officials "have not publicly confirmed whether the United States is developing or has developed LAWS" (IF11150, Mar 2026). Hegseth said the US "doesn't let AI make lethal targeting decisions" (AFR/Bloomberg, 1 May 2026). A DoD official said "a human is always in the loop for critical operational decisions" (Washington Sun/NOTUS, Jun 2026). Hegseth's 30 Sep 2026 speech created an Autonomous Warfare Command (target date 1 Oct 2027) and boasted of "thousands of autonomous systems" shipped. That is a label only, with no human-targeting or authorisation detail (speech, Agincourt memo). The Section 1066 and Section 1061 reports on autonomous-weapon approvals and waivers go to congressional committees and are not published.
• Russia. Ukrainian forensics and the NYT (Aug 2026) describe self-targeting Russian V2U and Molniya drones, but these are third-party claims and the Zaporizhzhia strike hit objects, not people (The Conversation, Sep 2026). Defence Minister Belousov has spoken only of AI for "image recognition, automatic target capture and navigation" (Jul 2026, via TAdviser). Russia and Ukraine both say a human stays in the loop, and "acknowledging full AI autonomy would provoke ethical and legal questions" (Al Jazeera, 14 Sep 2026).
• Israel, India, China, UK, France. I found no official claims. Israel's robotic systems are described as operator-controlled. Even Ukraine, which is not on the list and is the furthest ahead, officially denies fielding such systems (Small Wars Journal, Aug 2026).
• Diplomacy. The CCW expert group adopted a diluted consensus text in September 2026, and the Review Conference meets 16–20 Nov 2026 (Tech Times/HRW). Momentum toward negotiations makes explicit admissions more costly.

Base rate

Over about 12 years of debate on lethal autonomous weapons (2013–2026), none of the seven has made an official statement meeting this bar. That holds even though anti-materiel autonomy (Phalanx, C-RAM, Iron Dome) is described openly. The closest case is Russia's 2014–16 Ministry of Defence description of missile-site guard robots that "hit targets in automatic mode". It came during testing and was vague on authorisation. By analogy, armed-drone targeted killing took about 10 years from first use to US acknowledgment. Weapons that are stigmatised but not banned do eventually get acknowledged once they are militarily central.

Paths and hazard

• YES paths:
  • Russia's defence ministry, or Putin in a transcript on kremlin.ru, saying AI drones independently find and strike enemy "manpower". This is the most likely path, about 5–7% over five years.
  • US doctrine, testimony or releases after the 3000.09 rewrite and the stand-up of the Autonomous Warfare Command, or during a conflict (about 3%).
  • Israel or India describing automated border or installation defence (about 1–2% each).
  • China, UK or France (well under 1%).
  • Edge cases such as sensor-fuzed anti-personnel mines (about 1%).
• NO paths: governments keep "human in/on the loop" messaging and vague labels ("autonomous", "AI-enabled"), which do not qualify; details stay classified; treaty pressure; a Ukraine peace reduces battlefield boasting; and a possibly more cautious US administration after January 2029.

I model a quarterly hazard of about 0.4% now, rising to about 0.75% by 2031 as autonomy becomes normal. That is about 1.6% a year in 2027 and 2.9% in 2031. Cumulatively it gives about 2% by end-2027, 6% by end-2029 and 12% by end-2031 (computed in Python).

Strongest counter-case

Autonomy is spreading very fast, and the US and Russia removed human-oversight language from the CCW text. Over five years, perhaps with a Taiwan or other major war, explicit boasting could become routine, and a lenient reading of a Russian ministry post could resolve this early. I weighed that against the unbroken record of official denials and the demanding requirement that the source spell out how the system works and that its mode is authorised. That leaves me near 12% rather than 20% or more.

Question. By the last day of each quarter, has the government of the US, China, Russia, the UK, France, India, or Israel officially stated that it has fielded a weapon system authorised to select and attack human targets without a human approving each engagement?

Definitions.

• Government: the national executive government or its armed forces of any one of the seven named countries; China means the People's Republic of China. The statement must concern that government's own fielding and authorisation, not an allegation about another country.
• Fielded: delivered to and accepted by an operational military or security unit for actual operational use, including border or installation defence. Combat use is not required. Procurement, development, demonstrations, and deployments solely for testing or training are insufficient.
• Weapon system: a system that selects targets from sensor data and applies physical force against them, whether lethal or less-lethal. It need not use machine learning or be described as AI. Ordinary pressure-, contact-, tripwire-, or timer-triggered mines and booby traps do not constitute target selection; a system that detects and classifies people and decides which to attack does.
• Human targets: individual people or groups of people selected for attack as people, including enemy combatants. Attacking a vehicle, vessel, aircraft, radar, or building does not qualify solely because people are aboard or nearby.
• Authorised: the government's statement establishes that an operationally permitted mode of the fielded system allows this conduct. Technical capability alone, a prohibited mode, or a malfunction is insufficient. Authorisation may be limited to a location, mission, target class, or specified conditions.
• Without a human approving each engagement: after activation, the system can both choose the particular human target and initiate force against it without affirmative human approval of that target's engagement. Human mission planning, activation, supervision, and the ability to veto or abort do not disqualify it. Automatic tracking, aiming, or firing at a particular person or finite set of people whose engagement a human has already approved does not qualify.

Resolves YES if one or more qualifying official publications together explicitly establish fielding, permitted human-target selection, and attack without per-engagement human approval for the same system and operational mode. The system may be identified by name or by an unambiguous description. The concluding publication must fall within the Resolution window. A system described as already fielded before the opening date can qualify through a new official confirmation within that window. No actual attack, injury, or death is required.

Resolves NO if no qualifying confirmation has been published by the horizon. General doctrine allowing autonomous weapons, approval of development or future fielding, an order to revise policy, and labels such as “autonomous,” “fire-and-forget,” or “human-supervised” without the required functional and authorisation details do not qualify. Neither does a statement merely that no human intervened in a particular attack, unless it establishes that omission of per-engagement approval was authorised.

Resolving source. Official directives, doctrine documents, releases, published speeches, briefing transcripts, or formal government testimony or parliamentary answers from the relevant national executive, defence ministry, armed service, military command, or national security agency. Eligible authorities include the US President and Department of Defense/Department of War; China's State Council, Central Military Commission, Ministry of National Defense and PLA; Russia's Presidency and Ministry of Defence; the UK government and Ministry of Defence; France's Presidency, government and ministry responsible for the armed forces; India's government and Ministry of Defence; and Israel's government, Ministry of Defense and IDF, including successor bodies. Statements must be published on an official government or military channel, in an official legislative record, or as the government's formal submission in a UN document. Contractor claims, state-media reporting alone, unofficial remarks, leaks, and third-party assessments cannot resolve this question. This is a narrower official-government source requirement than Public confirmation (A2). Apply Resolution date (A3) and the Resolution window. Relevant policy documents such as DoD Directive 3000.09, January 25, 2023 and NSPM-11, June 5, 2026 do not alone establish fielding of a qualifying system.