The Western Consensus That Was Not: Resolution 80/57, the Rolling Text’s Human-Control Substitution, and the Structural Stakes of the November 2026 CCW Review Conference

[Hassan Khoder, LL.M., is a PhD candidate in Law at the University of Cologne in Germany specialising in autonomous cyber weapons and international humanitarian law. His working papers are available on SSRN.]

The Vote That Has Not Been Adequately Analysed

On 1 December 2025, the United Nations General Assembly adopted Resolution 80/57 on lethal autonomous weapons systems by 164 votes in favour to 6 against, with 7 abstentions. The six states voting against were Belarus, Burundi, the Democratic People’s Republic of Korea, Israel, the Russian Federation — and the United States. The United States had voted in favour of Resolution 78/241 (2023) and Resolution 79/62 (2024). Its reversal is the most significant diplomatic development in the autonomous weapons debate since the CCW Group of Governmental Experts was constituted in 2014. It has been noted. It has not been adequately analysed.

The prevailing reading treats the US vote as geopolitical defiance — a great-power recoil from multilateral constraint. That reading is not wrong, but it is incomplete. This piece advances a more specific argument: the US reversal is a rational legal response to a specific textual development in the CCW Chair’s rolling text — the deliberate substitution of ‘context-appropriate human judgement and control’ for ‘meaningful human control’ as the instrument’s governing standard. If retained in the November 2026 CCW Review Conference mandate, this substitution would expose operational US systems to a compliance standard the United States perceives as incompatible with DoD Directive 3000.09, as the US Explanation of Vote at the UNGA First Committee (1 November 2023) makes clear. The vote reveals not a rupture in Western consensus but its structural absence — a design flaw the November Review Conference must correct.

What the Rolling Text Actually Says

Part III of the rolling text is the operative core. It prohibits autonomous weapon systems that are ‘inherently indiscriminate’ or whose ‘effects in attack cannot be anticipated and limited’; requires ‘context-appropriate human judgement and control’; and enumerates enabling measures including limits on scale, target types, duration, and geographic scope, a prohibition on ‘significant’ self-modification through real-time machine learning, and timely deactivation mechanisms. Three observations are essential.

First, the rolling text’s prohibitions are structured as compliance failures, not architectural characteristics. A system ‘whose effects cannot be anticipated and limited’ is prohibited — but who determines whether effects can or cannot be anticipated is left to deploying states. The ICRC’s 2025 position paper identifies this flaw: a system ‘designed or used in a manner such that their effects cannot be sufficiently understood, predicted and explained’ is characterised as indiscriminate, yet neither paper nor rolling text specifies who makes that determination, by what standard, or through what verification mechanism. The prohibition is present; the compliance architecture is absent.

Second, ‘context-appropriate human judgement’ is not ‘meaningful human control’ with a synonym substituted. It is a meaningfully different standard. ‘Meaningful human control,’ as developed in the academic literature — see Peter Asaro’s foundational 2012 contribution in the International Review of the Red Cross — implies substantive cognitive engagement: a decision-maker who understands the targeting decision and bears genuine causal responsibility. ‘Context-appropriate’ introduces relativity: in high-tempo, machine-speed environments, the contextually appropriate level of human engagement may converge toward the vanishing point. Dorsey and Bo have demonstrated that automation bias in AI-enabled targeting workflows produces what they term the erosion of ‘human(e) judgment’ through quantification logics that reduce IHL assessments to algorithmic scores. The rolling text’s formulation opens space for precisely this erosion.

Third, the prohibition on ‘significant self-modification including through real-time machine learning’ is the provision whose operational implications the United States finds most constraining. The ‘significant’ qualifier is deliberately elastic, and whether the AI-enabled targeting architectures currently operating in active theatres satisfy it is determined by the deploying state, not by the text.

Why the United States Voted Against

DoD Directive 3000.09, reissued in January 2023, requires that autonomous weapons ‘are designed to allow commanders and operators to exercise appropriate levels of human judgment over the use of force.’ ‘Appropriate levels’ permits human judgment to be exercised at the level of system authorisation rather than individual engagement — accommodating advanced autonomous systems that operate without real-time human decision-making at each engagement. The rolling text’s Part III requirements approach a meaningful-control threshold without naming it. A United States prepared to litigate compliance within existing IHL is rationally unwilling to accept a new treaty standard whose content it does not control and whose operational implications it cannot manage.

The United States did not vote against Resolution 80/57 because it opposes civilian protection. It voted against because the rolling text’s prohibitions converge on a compliance architecture its operationally deployed systems may not satisfy, and it prefers existing IHL’s flexibility. The vote is a doctrinal signal about the human-control formulation’s compliance implications — and it should be read as such.

The Structural Flaw and the Verification Gap

The US vote reveals that the rolling text’s standard is structurally unstable: flexible enough to attract nominal support, specific enough to exclude the world’s most capable military power. The instrument will either be weakened to the point of deregulatory capture or adopted without the states most capable of generating the harms it targets.

The structural flaw runs deeper than the ‘context-appropriate’ formulation. The rolling text prohibits non-compliant systems but provides no verification architecture — no independent body, no technical standard for ‘anticipability,’ no pre-deployment certification — through which compliance could be assessed before deployment. The ICRC’s position paper recommends that states ‘put in place implementation measures’ — unspecified. Part IV of the rolling text contains an Article 36-style re-review obligation, but addressed to states, with no minimum procedural content and no consequence for non-performance.

This gap can only be closed by migrating the compliance locus upstream. The targeting obligations of Additional Protocol I — Articles 57(2)(a)(i), 51(5)(b), 48, and 52(2) — each presuppose a human decision-maker capable of independent legal assessment at the moment of engagement. The International Court of Justice characterised the principle of distinction as an ‘intransgressible principle’ of customary international law. For systems eliminating human verification at the moment of engagement by design, the compliance inquiry must migrate to the pre-deployment stage, where an Article 36 AP I weapons review is the only available mechanism.

What the November 2026 Review Conference Must Produce

The Review Conference must produce a negotiating mandate with three non-negotiable institutional elements absent from the current rolling text.

A mandatory declaration regime covering software and training data as well as hardware. The most dangerous autonomous targeting systems are defined by their algorithms; a declaration regime limited to hardware misses the weapon.

A challenge inspection mechanism with a 72-hour activation timeline providing access to source code, training data, and operational logs. The OPCW’s challenge inspection mechanism has been formally invoked only once — by the Netherlands against Russia in 2018 — precisely because its existence deters the violations it would detect. The same deterrent logic applies here.

A pre-deployment certification requirement preventing deployment until an independent Technical Secretariat — staffed with AI-system auditors, IHL legal experts, and verification specialists, modelled on the OPCW’s — certifies that the system’s design parameters satisfy the applicable prohibitions. The full institutional design of this framework — the International Autonomous Weapons Compliance Organization (IAWCO) — is set out in the author’s working papers on SSRN (Abstract No. 7182838 and Abstract No. 7182839).

The Argument Whose Time Has Come

The ICRC has recommended — correctly — that states prohibit autonomous weapon systems ‘designed or used in a manner such that their effects cannot be sufficiently understood, predicted and explained.’ The rolling text echoes this. But both stop at the prohibitory norm, leaving compliance assessment to the states the norm constrains. The result is a regime that declares principles while the systems it targets proliferate: twenty thousand active Maven Smart System users across every US combatant command; AI-enabled targeting cells in multiple active conflicts; the ICRC identifying automation bias as a field-level risk in sixteen formal recommendations to the UN Secretary-General (April 2025).

The US ‘no’ vote on Resolution 80/57 is a forecast. It forecasts a coalition of over seventy states — including a growing group of African states, as reported by Reaching Critical Will— pressing for negotiations, against a small but militarily decisive bloc resistant to compliance accountability. A mandate retaining ‘context-appropriate’ without the three institutional elements above will produce an instrument that governs nothing its text names.

The Ottawa Process produced the Mine Ban Treaty in fourteen months. The Oslo Process produced the Convention on Cluster Munitions in two years. Both succeeded because a critical mass of states decided the humanitarian cost of inaction exceeded the military cost of legal constraint — and because both produced specific treaty texts, not statements of principle. The legal architecture for such a text on autonomous weapon systems already exists. The November 2026 CCW Review Conference can establish, irrevocably, that declaration, challenge inspection, and pre-deployment certification are the minimum architecture for any instrument that aspires to govern the systems it names. The legal reasoning is available. The political decision is what remains.

Photo attribution: “Responsible AI in the Military Domain – REAIM 2023” by Ministerie van Buitenlandse Zaken is licensed under CC BY-SA 2.0