From 16 to 20 November 2026 the Seventh Review Conference of the Convention on Certain Conventional Weapons will decide what to do with a consensus report that characterises lethal autonomous weapon systems and sets out elements for a possible instrument, without prejudging its legal nature. The Group of Governmental Experts completed its 2024-2026 mandate on 4 September 2026. Its conclusions affirm that international humanitarian law applies fully, that human judgement and control are required for compliance, and that accountability cannot be transferred to machines. They do not yet create a new protocol, a prohibition, or a verification regime. This paper reads the characterisation and the recommended elements against the Convention's structure, examines the strongest objections to both premature legalisation and indefinite delay, and sets observable tests for judging the Review Conference without mistaking endorsement of elements for operational restraint.
Key findings and scope
- The Group of Governmental Experts adopted its final report on 4 September 2026 as CCW/GGE.1/2026/3 (advance copy dated 14 September 2026). It submits a consensus set of elements in paragraphs 25-44 and recommends that the Seventh Review Conference consider and endorse them without prejudice to future discussions [1].
- Within the scope of the Convention, a lethal autonomous weapon system is characterised as a combination of one or more weapons and technological components operating together that is capable of identifying, selecting and engaging a target in a way that results in loss of life, without intervention by a human being operating the system. Programming a set of potential targets does not take a system outside that characterisation; determining a specific target and the timing of an attack does [1, paragraph 25].
- The report states that international humanitarian law applies fully to the development and use of such systems, that control and human judgement are needed to uphold distinction, proportionality and precautions in attack, and that accountability cannot be transferred to machines [1, paragraphs 29-41].
- The Review Conference, scheduled for 16-20 November 2026 in Geneva and to be presided by Switzerland, must decide by consensus on possible modalities of future Group work. Estimated-cost documents for a 2027 Group session are already listed among conference papers [2; 3].
- Italy's Permanent Mission to the Conference on Disarmament reports 128 High Contracting Parties and records that at least seventy of them have expressed readiness to move toward negotiations on a legally binding instrument. Those figures are diplomatic reporting of political positions, not a vote tally of the Review Conference [4].
This research uses documents available by 6 November 2026. Treaty and adopted conference texts establish the legal baseline; the Group's final report establishes consensus negotiating results to date; meeting pages establish the calendar. Side-event programmes and civil-society commentaries are treated as evidence of debate, not as evidence of governmental agreement. No judgement is offered on whether any particular deployed or developmental system meets the Group's characterisation, because the public evidence assembled for this paper cannot support one.
1. Why the Convention is the venue, and what that constrains
The Convention on Prohibitions or Restrictions on the Use of Certain Conventional Weapons Which May Be Deemed to Be Excessively Injurious or to Have Indiscriminate Effects is a framework treaty. Protocols attached to it address specific categories of weapons. That architecture matters for autonomous systems. A new protocol could, in principle, prohibit or restrict a defined class of weapons. A political declaration or a set of endorsed elements could clarify existing legal obligations without creating new ones. The Group's mandate, renewed by the 2023 Meeting of High Contracting Parties, asked it to formulate by consensus a set of elements of an instrument, without prejudging its nature, and other possible measures, taking into account existing Protocols, proposals by High Contracting Parties and other options [1, paragraph 1].
The phrase "without prejudging its nature" is not diplomatic filler. It kept States that want a binding protocol and States that prefer soft guidance inside a single drafting exercise. It also means that consensus on elements is not consensus on a Protocol. The Review Conference can endorse the elements, mandate further work, open negotiations, or do some combination. It cannot treat the September report as if the legal instrument had already been chosen.
The Group met for ten days in each of 2024, 2025 and 2026, with additional informal meetings. In 2026 the sessions ran from 2 to 6 March and from 31 August to 4 September. The Chair throughout the 2024-2026 cycle was Ambassador Robert in den Bosch of the Kingdom of the Netherlands, after the Eastern European Group forwent its turn and High Contracting Parties endorsed the Western European and Others Group nomination under a written silence procedure [1, paragraphs 2-5; 3]. Continuity of chairing helped, but it did not remove the requirement of consensus for the report itself.
2. Reading the characterisation before the headlines
Paragraph 25 is the hinge of the report. A lethal autonomous weapon system, within the scope of the Convention, can be characterised as a combination of one or more weapons and technological components operating together, capable of identifying, selecting and engaging a target in a way that results in loss of life, without intervention by a human being operating the system [1, paragraph 25]. Three clarifying points follow.
First, the fact that a human being programs or otherwise determines a set of potential targets or characteristics thereof does not exclude the system from the characterisation. A system is not included when personnel planning, deciding upon or carrying out the attack determine the specific target or specific group of targets and the timing of the attack [1, paragraph 25(a)]. The distinction is between setting parameters in advance and selecting the concrete object of attack at the moment of engagement. That is a legal and operational distinction, not a marketing one.
Second, the fact that such a system can be used in a way that does not result in loss of life, for example to damage objects or cause injury, does not exclude it from the characterisation [1, paragraph 25(b)]. The characterisation is therefore not limited to systems used only for killing. It is framed around capability that can produce loss of life.
Third, the characterisation is without prejudice to future understanding and to the potential modification of the characterisation, as well as to the exclusion or inclusion of certain types of systems [1, paragraph 25(c)]. Endorsement of the present text would therefore not freeze technology categories forever. It would establish a shared baseline from which later instruments could tighten, widen or refine the definition.
Meridian's assessment is that paragraph 25 is more precise than most public summaries suggest and less conclusive than some advocacy claims imply. It supplies a workable object for regulation. It does not by itself prohibit any system, require any particular form of human approval at the moment of attack, or settle dual-use questions about autonomy in intelligence, logistics or electronic warfare.
3. What the elements require of humans, and what they leave open
International humanitarian law as the floor
The report states that international humanitarian law applies fully to the development and use of lethal autonomous weapon systems, and that the right of parties to choose methods and means of warfare is not unlimited [1, paragraphs 29-30]. Systems must not be used if they are of a nature to cause superfluous injury or unnecessary suffering, or if they are inherently indiscriminate, including if they cannot be directed at a specific military objective or if their effects cannot be limited as required by international humanitarian law and consequently are of a nature to strike military objectives and civilians or civilian objects without distinction [1, paragraph 31]. Attacks must not be directed against the civilian population as such, civilian objects or individual civilians protected under that law [1, paragraph 32]. Use must comply with distinction, proportionality and precautions in attack [1, paragraph 33].
These restatements matter because some public debate treats autonomy as if it created a legal vacuum. The Group rejects that view. Restating the floor, however, is not the same as specifying how machine perception, target recognition and engagement timing can satisfy distinction and proportionality under combat conditions. That specification is where an instrument would have to go further than a restatement.
Control, judgement and the chain of command
Paragraph 34 is the second hinge. To uphold compliance with distinction, proportionality and precautions, control and human judgement with regard to such systems are needed. Human beings exercise such control directly or indirectly, including through measures taken before and/or during use. Regardless of how control is exercised, human beings need to exercise judgement to determine the measures necessary to ensure compliance and to implement those measures in practice, considering operational context and the characteristics and capabilities of the weapon system [1, paragraph 34].
Relevant measures include, among others: operation under a responsible chain of human command and control; limiting the scale of the operation, types of targets, duration and geographical scope, as appropriate; and measures to intervene, to the extent possible or feasible, including to deactivate or neutralize systems in a timely manner [1, paragraph 35]. The language is careful. "To the extent possible or feasible" acknowledges that intervention may not always be available once a system is committed. Indirect control through earlier design, testing and mission constraints is expressly contemplated. That breadth is both a strength, because it covers real military practice, and a vulnerability, because a State could claim that pre-mission parameter-setting alone satisfies the requirement.
Life-cycle duties, bias and accountability
In the study, development, acquisition or adoption of a new system, States must determine whether its employment would, in some or all circumstances, be prohibited by international law. Voluntary exchange of practices is encouraged, bearing in mind national security and commercial restrictions [1, paragraph 36]. During design, development, testing and deployment, risks of civilian casualties and precautions to help minimise incidental harm must be considered, together with risks of unintended engagements, loss of control, proliferation and acquisition by terrorist groups [1, paragraph 37].
States should ensure that capabilities and limitations can be understood through testing and evaluation, including adequately realistic simulations, and through guidance, training and instructions for personnel in the chain of command and control [1, paragraph 38]. They should implement measures to prevent and mitigate unintended bias in data sets and artificial intelligence models, and automation bias in use [1, paragraph 39]. States, parties to armed conflict and individuals remain at all times responsible under applicable international law for their decisions regarding use. Accountability cannot be transferred to machines. States must take measures to ensure individual accountability, including through operation within a responsible chain of human command and control, and should promote human responsibility through understandable human-machine interfaces, guidance consistent with international humanitarian law, and training [1, paragraphs 40-42]. They must take measures necessary to prevent and suppress violations under their jurisdiction and are encouraged to establish effective internal mechanisms for addressing incidents that may involve such violations [1, paragraphs 43-44].
Table 1. What the September elements settle and what they leave to the Review Conference
| Issue | Settled in the Group report (consensus text) | Still open for November |
|---|---|---|
| Object of discussion | Characterisation in paragraph 25, with carve-outs for specific-target human decisions | Whether to modify, freeze or build a Protocol around that characterisation |
| Legal floor | Full application of international humanitarian law; no transfer of accountability to machines | How to operationalise compliance tests for machine perception and engagement timing |
| Human role | Control and human judgement required; direct or indirect; chain of command | How much real-time intervention is required, and how "feasible" is interpreted |
| Nature of instrument | Explicitly not prejudged | Protocol, political declaration, combination, or further mandate only |
| Future work | Recommendation to decide modalities by consensus | Duration, negotiating authority, relationship to other forums |
Source: CCW/GGE.1/2026/3 [1], paragraphs 25-45. Open items are Meridian analysis of the recommendation text; they are not scored as failures of the Group.
4. The Review Conference calendar and the decision that actually matters
The Seventh Review Conference will meet from 16 to 20 November 2026 at the Palais des Nations in Geneva, under the presidency of Switzerland's Permanent Representative to the Conference on Disarmament. An aide-mémoire for participants was circulated on 9 October 2026. Speaking-time limits for the general exchange of views are five minutes per delegation; three minutes apply for other agenda items [2]. Those limits are procedural, but they signal a crowded agenda in which lethal autonomy will compete with universality, finance, Implementation Support Unit reporting and the rest of the Convention's machinery.
Conference documents already listed on the United Nations Office for Disarmament Affairs meeting page include provisional agendas and programmes for the Conference and its Main Committees, reports on the sponsorship programme, universality, the Implementation Support Unit and the status of financial contributions, estimated costs for a 2027 Group of Governmental Experts on emerging technologies in the area of lethal autonomous weapon systems, and a proposal on the mandate and modalities of the future work of that Group [2]. The existence of a 2027 cost estimate does not decide that the Group continues in its present form. It does show that continuation is among the live options being costed.
The Group's own recommendation is twofold: that High Contracting Parties consider and endorse the set of elements in paragraphs 25-44 without prejudice to the result of future discussions; and that the Review Conference decide by consensus on the possible modalities of future Group work [1, paragraph 45]. Endorsement without a negotiating mandate would leave the elements as a political and legal reference point. A negotiating mandate without endorsement would reopen characterisation. Both together would be the strongest incremental outcome short of opening a Protocol drafting process. None of those outcomes, by itself, would place inspectors in factories or constrain systems already fielded under national interpretations.
5. The strongest objections, and where they hold
Objection 1: consensus elements are a substitute for prohibition
The strongest critical case is that a characterisation-plus-elements package allows major military powers to claim progress while preserving wide discretion over autonomy in targeting. On this view, "without prejudging its nature" was a device to avoid a ban, and "indirect" control dilutes meaningful human judgement into paperwork completed months before an engagement. The concern is serious. Paragraph 25(a) does treat prior parameter-setting as compatible with autonomy, and paragraph 35's intervention language is qualified by feasibility [1]. Meridian's assessment is that the objection correctly identifies a risk of hollow compliance. It does not follow that the elements are worthless. A shared characterisation and an explicit statement that accountability cannot be transferred to machines are necessary conditions for later restriction. They are not sufficient conditions for restraint in war.
Objection 2: premature legalisation would freeze the wrong technology
The strongest sceptical case from governments wary of new rules is that binding definitions adopted now would either be overtaken by machine learning advances or would capture systems that remain under effective human control in practice. The United States working paper listed among Review Conference documents, and national statements in related forums, reflect variants of this caution, though Meridian has not treated secondary press summaries of those positions as primary evidence of the Conference outcome [2]. The Group's own paragraph 25(c), preserving room to modify the characterisation, is a partial answer [1]. It is only a partial answer: amendment of a future Protocol would still require consensus.
Objection 3: the Convention is the wrong forum
A further objection is that the Convention's consensus culture and membership cannot produce timely rules for a technology already entering inventories, and that other forums or national regulation should lead. Italy's Mission reports that a majority of High Contracting Parties among those supporting negotiations are ready to move to a legally binding instrument, while also recording that consensus was reached on elements only after difficult overnight negotiation in early September [4]. The report itself shows broad participation: 103 High Contracting Parties' delegations are listed as participants in the Group's work, together with two Signatories and fifteen States not party as observers [1, paragraphs 13-15]. Breadth of participation strengthens legitimacy. It does not guarantee speed. Meridian's judgement is that abandoning the Convention track would trade a forum that includes major military powers for processes that may move faster on paper and bind fewer of the States that matter most for deployment. That trade-off should be stated honestly rather than denied.
Objection 4: focusing on weapons ignores the wider military artificial intelligence stack
Autonomous weapons sit inside a wider set of artificial intelligence applications for intelligence, planning, logistics and decision support. A weapons-only instrument could miss the upstream systems that shape targeting. The Group's mandate, however, was weapons-focused within the Convention. Demanding that one Group solve the entire military artificial intelligence problem is a way of ensuring it solves none of it. The better test is whether the Review Conference preserves coherent links between weapons rules, Article 36-style legal reviews of new means and methods of warfare, and national accountability systems. The September text already points toward life-cycle legal review and internal incident mechanisms [1, paragraphs 36-44]. Those pointers need national practice, not only conference language.
6. Observable tests for 16-20 November
Meridian proposes five tests that can be applied to the Review Conference outcome without requiring speculation about classified programmes.
- Characterisation: Does the final declaration or decision endorse paragraph 25 substantially intact, amend it, or sidestep endorsement?
- Nature of instrument: Does the Conference open negotiations on a Protocol, mandate further element-drafting, adopt a political declaration only, or defer?
- Human judgement: Do any agreed modalities specify whether indirect pre-use measures can satisfy the control requirement in the absence of real-time intervention capacity?
- Resources: Are assessed contributions and sponsorship arrangements sufficient for whatever mandate is agreed, or does the Conference create an underfunded process?
- Reporting: Does the outcome create any expectation of transparency about national reviews of new systems, beyond voluntary practice exchange?
A Conference that endorses the elements, agrees a time-bound negotiating mandate and funds participation would be a material step. A Conference that only renews open-ended discussion would preserve the September baseline but would not convert it into rules. A Conference that fails to agree modalities by consensus would leave the elements as a completed Group product without a clear custodial process. All three are possible on the documents as they stand.
7. Conclusion
The September report is a real diplomatic achievement in a polarised security environment: a consensus characterisation, a restated humanitarian-law floor, and an explicit refusal to transfer accountability to machines. It is not a treaty, not a ban, and not a verification system. The Seventh Review Conference will show whether High Contracting Parties treat those elements as a ceiling, a floor for negotiation, or a text to be quietly shelved. Outside observers should judge November by the five tests above, not by the volume of rhetoric about innovation or about killer robots. The difference between endorsed elements and operational restraint will be settled in national acquisition, targeting directives and legal reviews long after the Palais des Nations empties.
References
- Group of Governmental Experts on Emerging Technologies in the Area of Lethal Autonomous Weapons Systems, Report of the 2024-2025-2026 sessions, CCW/GGE.1/2026/3, advance copy, 14 September 2026. https://italiarappdisarmo.esteri.it/wp-content/uploads/2026/09/CCW-GGE.1-2026-3-Advance-copy.pdf (accessed 6 November 2026).
- United Nations Office for Disarmament Affairs, "Convention on Certain Conventional Weapons, Seventh Review Conference (2026)", Meetings Place, https://meetings.unoda.org/ccw-revcon/convention-on-certain-conventional-weapons-seventh-review-conference-2026 (accessed 6 November 2026).
- United Nations Office for Disarmament Affairs, "Convention on Certain Conventional Weapons, Group of Governmental Experts on Lethal Autonomous Weapons Systems (2026)", Meetings Place, https://meetings.unoda.org/ccw-/convention-on-certain-conventional-weapons-group-of-governmental-experts-on-lethal-autonomous-weapons-systems-2026 (accessed 6 November 2026).
- Permanent Mission of Italy to the Conference on Disarmament, "The 128 States Parties to the Convention on Certain Conventional Weapons (CCW) have agreed on the basic elements for defining lethal autonomous weapons and their use", 5 September 2026 (page date), https://italiarappdisarmo.esteri.it/en/news/dalla_rappresentanza/2026/09/the-128-states-parties-to-the-convention-on-certain-conventional-weapons-ccw-have-agreed-on-the-basic-elements-for-defining-lethal-autonomous-weapons-and-their-use/ (accessed 6 November 2026).
- United Nations Office for Disarmament Affairs, Convention on Certain Conventional Weapons overview page, https://www.unoda.org/en/our-work/conventional-arms/convention-certain-conventional-weapons (accessed 6 November 2026).
- Indico.UN, "7th Review Conference of the High Contracting Parties (16-20 November 2026): Overview", https://indico.un.org/event/1019366/ (accessed 6 November 2026).
- United Nations Office at Geneva, event notice for the Seventh Review Conference of the High Contracting Parties to the CCW, https://www.ungeneva.org/en/meetings-events/events/observances-special-events/2026/11/seventh-review-conference-high (accessed 6 November 2026).
- Meeting of the High Contracting Parties to the CCW, final report CCW/MSP/2023/7 (mandate language as reproduced in CCW/GGE.1/2026/3, paragraph 1) (accessed via [1], 6 November 2026).
- Meeting of the High Contracting Parties to the CCW, CCW/MSP/2025/8, paragraph 19 (2026 session dates), as cited in [3] and [1] (accessed 6 November 2026).
- International Committee of the Red Cross, working paper CCW/GGE.1/2026/WP.4, listed on the 2026 Group meetings page [3] (document listing verified 6 November 2026; substantive claims in this paper rely on the adopted Group report rather than the working paper's full text).
- Opinio Juris symposium introduction, "The End of the Beginning", 5 October 2026, used only as secondary confirmation of the overnight consensus timing and of reporting that at least seventy High Contracting Parties support negotiations; not used as authority for legal characterisation, https://opiniojuris.org/2026/10/05/symposium-on-the-ccw-gge-laws-the-end-of-the-beginning-an-introduction/ (accessed 6 November 2026).
- Human Rights Watch, "UN Talks on Killer Robots End with Calls for Negotiations Growing", 7 September 2026, used only for civil-society count of States supporting negotiations (76 total / 70 High Contracting Parties as reported by Stop Killer Robots via HRW); legal analysis rests on [1], https://www.hrw.org/news/2026/09/07/un-talks-on-killer-robots-end-with-calls-for-negotiations-growing (accessed 6 November 2026).
- United Nations Office for Disarmament Affairs, CCW Sponsorship Programme and Implementation Support Unit report listings under Review Conference documents [2] (accessed 6 November 2026).
- CCW/CONF.VII/7 and CCW/CONF.VII/8 estimated-cost documents for 2027 Group and Meeting/ISU costs, listed on [2] (listings verified 6 November 2026).
- Proposal on the mandate and modalities of the future work of the GGE on LAWS, Main Committee II document listed on [2] (listing verified 6 November 2026).
- Sixth Review Conference rules of procedure CCW/CONF.VI/11, as applied mutatis mutandis to the Group per [1], paragraphs 6-11 (accessed via [1], 6 November 2026).
- Stop Killer Robots side-event listings on the 2026 Group meetings page [3], used only as evidence that negotiating and procurement questions were debated publicly during the mandate (accessed 6 November 2026).
- UNIDIR scenario-based workshop listings on the 2026 Group meetings page [3], used only as evidence of IHL scenario work offered to delegates, not as an independent legal opinion (accessed 6 November 2026).