Autonomous Weapons Systems Mini-Symposium: Crunch Time for the Discussion About Autonomous Weapons

[Rain Liivoja is a Professor at the University of Queensland Law School and a Director of the Asia-Pacific Institute for Law and Security.]

Editorial Note: This post forms the first of three posts of the Opinio Juris mini-symposium on the upcoming meeting of the Group of Governmental Experts on Lethal Autonomous Weapons Systems within the auspices of the Convention on Certain Conventional Weapons.

A version of these reflections was delivered at the ASEAN Regional Forum Inter-Sessional Meeting on Non-Proliferation and Disarmament in April 2026.

The long-running multilateral discussion about the regulation of autonomous weapon systems is entering a critical phase. On 31 August 2026, the Group of Governmental Experts (GGE) on Lethal Autonomous Weapon Systems (LAWS) will start its final week of work under the current three-year mandate to “formulate, by consensus, a set of elements of an instrument, without prejudging its nature, … to address emerging technologies in the area of lethal autonomous weapon systems”.

The Group has been working towards fulfilling this mandate by iterating a “rolling text”, the most recent version of which was released on 5 June 2026. This text represents a significant effort by the Chair and participants in the GGE to develop a tangible normative outcome. It draws on existing international law and on the Group’s previous work (including its 2018/2019 Guiding Principles). It reflects the broad agreement on the two-tier regulatory approach whereby some LAWS are deemed prohibited due to incompatibility with international humanitarian law (IHL), while others are subject to regulation. It covers a range of issues, including the characterisation of LAWS; the application of IHL (including specific prohibitions and the residual Martens Clause); the need for, and modalities of, human–machine interaction; governance measures to be taken throughout the life cycle of a LAWS (including legal review); and accountability and responsibility.

The Chair has proposed incorporating the finalised set of elements from the rolling text into the Group’s final report. The Group would, under this proposal, invite the Seventh Review Conference of the Convention on Certain Conventional Weapons (CCW) in November 2026 to endorse the elements and to decide, by consensus, on next steps.

This post flags three central issues that would need to be resolved in order for the set of elements in the Chair’s draft report to attract the widest possible support within the GGE on the evening of 4 September — or, if the marathon meeting at which the Guiding Principles were adopted is anything to go by, the early morning of 5 September. These issues relate to the characterisation of LAWS, human–machine interaction, and certain responsible use principles (sometimes dubbed ‘-ilities’).

Characterisation

The characterisation of LAWS is where progress over the past two years has been most difficult but perhaps also most striking. The rolling text has clearly leaned towards a broad conceptualisation, which brings in scope all weapon systems that “can identify, select, and engage a target, without intervention by a human operating the system”. This has been facilitated in part by a convergence on a two-tier regulatory approach, which does not seek to prohibit all weapon systems that may be characterised as LAWS.

Some uncertainties remain about the best way to unpack the notion of a “weapon system” (this term having caused difficulty for some delegations). The current language — “a combination of one or more weapons and functionally integrated technological components” — is rather clunky but seems to get the job done. Also, some further discussion may arise from reflecting lethality and the associated saving clause preventing LAWS that are capable of non-lethal uses from falling out of scope.

That said, the decisive issue seems to be the inclusion of the identification of targets, in addition to selection and engagement, as one of the critical functions of a LAWS. This gave rise to concerns about unduly narrowing the characterisation, for example because it might capture only systems that can develop their own target profiles, and exclude the more likely case of systems that match persons or objects encountered in the environment against a pre-set target profile. The solution was to include a clause that explains that a system can be a LAWS even if “a human being designs, programs, or otherwise determines a set of potential targets or characteristics thereof”. Conversely, the text now also clarifies, with a view to excluding precision-guided munitions and the like from the characterisation of LAWS, that a system is not a LAWS “when the personnel planning, deciding upon, or carrying out the attack determines the specific target or group of targets to be engaged by the system”. The question is whether this two-pronged approach and the specific language sufficiently address concerns about potential over- or underinclusiveness of the characterisation.

Human Intervention

Aside from characterisation, the most persistent disagreements in the GGE have clustered around the way in which the elements of an instrument should reflect legally and ethically desirable human–machine interaction in the use of LAWS that are not prohibited.

Throughout much of the GGE debate, many States and civil society actors have promoted a requirement of “meaningful human control”. But a sizeable group of States rejected this notion outright. Some put forward alternative formulations such as “appropriate levels of human judgement” or “context-appropriate human involvement”. This led to the slightly convoluted “context-appropriate human judgement and control”, but this has now been pared back to the more straightforward “human judgement and control”. States no doubt have strong views as to whether this is appropriate or adequate.

There has also been a challenge around the role that this human intervention requirement, however phrased, would serve in an instrument. Some participants have regarded a specified degree of human intervention as an objective in its own right, linking it to ethical or moral notions such as human dignity. Others have treated human intervention as (mostly) instrumental: something necessary to ensure that the use of LAWS complies with existing customary and treaty IHL. 

The latter approach has prevailed in the rolling text. The current version makes this clear by stipulating that “[t]o uphold compliance with international law, in particular IHL, … human judgement and control with regard to LAWS are needed.” The text then clarifies that “[t]his does not imply constant direct human control” but that, “in determining the appropriate level of human judgement and control needed, a range of factors should be considered, including the operational context and the characteristics and capabilities of the weapon system as a whole”. 

The text goes on to provide an indicative list of “measures” through which such judgement and control can be, or ought to be, “implemented”. The language here remains quite complicated and would certainly benefit from further refinement — which many participants will no doubt attempt. 

One of the reasons for the persistent difficulties with human control may be the tendency to treat it as an ontologically distinct state of being — a special condition that is achieved through a range of practical measures and that then ensures compliance with legal rules or moral imperatives. Perhaps a better way would be to think of human control simply as a shorthand for the various measures taken to achieve operational effectiveness, legal compliance and ethical alignment. On this view, the measures listed in the rolling text are not the means by which human control is achieved or implemented, but simply elements constituting human control. 

This approach might allow for a simpler formulation to the effect that human control and judgement must be exercised through the taking of certain measures. It would also be consistent with the sensible recommendation that States should focus their efforts on determining how the various measures of human intervention apply in practice (rather than, say, wordsmithing an abstract statement about human control). In any event, the precise articulation of desirable human interaction will be one of the key issues for the GGE’s final week.

The ‘-ilities’

Over the years, the GGE has discussed multiple principles that have become common in AI governance frameworks. These notably include reliability, explainability and traceability. With the strong endorsement of the International Committee of the Red Cross, the discrete requirement of predictability has also received a lot of airtime — a distinctive feature of the GGE debates as compared to, for example, discussions about the governance of military AI.

The GGE has struggled to reflect these principles in previous consensus-based reports, largely because several delegations have found them unclear or insufficiently grounded in existing law. Multiple iterations of the rolling text have, however, included the proposition that, in order for LAWS to be used lawfully, “their effects must be adequately predictable, reliable, traceable and explainable”. But as the concepts have remained undefined, this language has also attracted the criticism that it lacks certainty. The underexplored interaction between the principles and existing law also does not help.

The current rolling text takes a different tack. It addresses this issue in the broader context of the life cycle of a LAWS. Specifically, according to the rolling text, States should ensure that the anticipated effects of LAWS are understood, which includes ensuring that LAWS “in their identification, selection, and engagement functions, are adequately predictable and reliable in operation, and that their functioning and effects are traceable and explainable to those responsible for their development and use”.

The additional detail may be helpful in elucidating the meaning of the principles but may equally raise additional questions. For example, what is the consequence of predictability no longer referring to effects, but to LAWS (or their functioning) more generally? Moreover, the new language is unlikely to satisfy all delegations. 

A more fruitful approach might be to reflect the substantive requirements of predictability, reliability, traceability and explainability across the various elements of the instrument, rather than condensing them, as labels, into a single element. For example, the proposed requirement to “ensure that the capabilities and limitations of LAWS in their expected use and their anticipated effects in different circumstances are understood” might be an adequate reflection of explainability. Likewise, the prohibition of the use of LAWS, “if their effects in attack cannot be anticipated and limited, as required by IHL in the circumstances of their use”, seems to capture much of predictability.

Conclusion

Statements by States reflect the maturity of the rolling text, even in its previous iterations. In September 2025, 39 High Contracting Parties to the CCW delivered a joint statement, recording their view that the text, while a work in progress, constitutes a sufficient basis for negotiations on an instrument, and declaring themselves ready to move towards such negotiations within the CCW. In March 2026, they were joined by the African States that are High Contracting Parties to the CCW, bringing the number of States supporting negotiations on the basis of the rolling text to over 70. 

For the time being, however, the focus must be on fine-tuning the text and, one hopes, adopting some version of it in early September. But therein also lies a major challenge: the search for consensus should not allow the existing text to be whittled down to mere platitudes that add nothing to the existing governance framework. Simply reciting existing law and reiterating consensus language from several years ago would be inadequate in light of the rapid development of the technology.

There is disagreement over whether the rolling text includes, and should include, language that goes beyond the existing law. Whatever the case may be, the rolling text certainly articulates practical measures that would enhance legal compliance and promote the responsible use of technology. The removal of these elements, for example on the logic that States have discretion in the choice of domestic IHL implementation mechanisms, would undo years of painstaking work within the GGE. 

Photo attribution: “View of the United Nations Office in Geneva adorned with flags of various countries” by Hugo Magalhaes