2026 Group of Governmental Experts on Lethal Autonomous Weapon Systems
Second Session (Geneva, 30 August–4 September 2026)
Asia-Pacific Institute for Law and Security
Utrecht University
Proposed text
39. States must, in accordance with their obligations under international law:
a. Conduct legal reviews in the study, development, acquisition or adoption of LAWS to determine whether their employment would, in some or all circumstances, be prohibited by international law. In this context, the voluntary exchange of relevant practices between States is encouraged, bearing in mind national security considerations and commercial restrictions on proprietary information;
b. Ensure that any further development of, or modifications to, existing LAWS that have a substantial impact on the operational functioning of their target identification, selection, and engagement functions, or on their anticipated effects, are reviewed to ensure continued compliance with international law.
40. States should, as appropriate, across the life cycle of LAWS, ensure that the capabilities and limitations of LAWS in their expected use and their anticipated effects in different circumstances are understood by:
a. Conducting testing and evaluation that can assess their performance within representative and anticipated operational environments, including by means of simulations; …
47. States must ensure effective internal mechanisms for the investigation, reporting, and appropriate action in response to incidents involving their use of LAWS that may involve violations of IHL. States should retain records, such as operational logs, sufficient to enable them to recount and review the functioning of LAWS.
Statement as delivered
Thank you, chair!
I am delivering this intervention on behalf of the delegation of the Asia-Pacific Institute for Law and Security (APILS), and my own delegation, that of Utrecht University.
Paragraphs 39
We welcome the revised language concerning legal reviews in paragraph 39. The paragraph now consolidates the text on legal reviews, and separates it from other measures, mentioned in subsequent paragraphs, taken across the life cycle of LAWS. We believe this separation is appropriate as the majority of States are under a legal obligation to carry out legal reviews. The use of the word “must” in conjunction with the phrase “in accordance with their obligations under international law” in the chapeau acknowledges that many States are under a direct legal obligation to conduct reviews.
We welcome the introduction of language mirroring this specific legal obligation in subparagraph (a). This has, however, made some of the language in the chapeau redundant. The reference to the study, development, acquisition and adoption of LAWS in subparagraph (a), and the reference to development and modification of LAWS in subparagraph (b), already indicate where in the life cycle of LAWS it is appropriate to conduct legal reviews. Therefore, we suggest deleting from the chapeau the words “as appropriate, across the life cycle of LAWS, and”. The chapeau would then simply read:
States must, in accordance with their obligations under international law:
We consider subparagraph (b) to be a valuable contribution of this Group to the evolving practice of legal reviews. It articulates a need to conduct further legal reviews when changes to existing weapon systems result in substantial changes to their functioning.
In this subparagraph we suggest inserting the word “continued” before the word “compliance” to further emphasise the life cycle approach to ensuring legal compliance. We also suggest replacing the reference to “IHL” at the end of the subparagraph with “international law” to bring it in line subparagraph (a). So the final clause of subparagraph (b) would read:
are reviewed to ensure continued compliance with international law.
Paragraphs 40
In paragraph 40, subparagraph (a), explains what simulations are intended to achieve. In our view, this subparagraph should indicate the objective of testing and evaluation more generally, and then identify simulation as one possible method of testing and evaluation. This can be achieved by minor restructuring.
We are also concerned that the deletion of the word “variable” before “environments” unduly narrows the testing requirement. It would tolerate testing with respect to a limited range of environments that do not reflect the actual operational uses of the LAWS. Thus we would suggest retaining the word “variable” or replacing it with “representative and anticipated”.
The subparagraph could then read:
Conducting testing and evaluation that can assess their performance within representative and anticipated operational environments, including by means of simulations;
Chair,
Delegations hold diverging views about subparagraph (c). It contains a number of concepts that many delegations have recognised as important. Some of these concepts have also been reflected in national and international documents relating to the use of military artificial intelligence.
Some delegations have, however, suggested that these concepts do not relate to existing IHL obligations. We do not think this is the case. For example, predictability does not appear in IHL instruments as a discrete legal requirement. But, as an implicit operational necessity, predictability underpins compliance with a number of principles and rules of IHL. This includes the duty to take feasible precautions and to assess the proportionality of collateral damage, as these presume the ability to foresee the effects of attacks.
Other delegations have, in previous occasions, noted the lack of certainty about the meaning of predictability, reliability, traceability and explainability. This is a valid concern. Therefore, we believe that incorporating the substantive requirements of these notions across the various elements of the instrument is more fruitful than condensing them, as labels, into a single subparagraph here.
We also believe that the text in many respects already encapsulates the notions of predictability, reliability and explainability.
Predictability is reflected in paragraph 32, which contains the prohibition to use LAWS the effects of which cannot be anticipated. There were further references to anticipated effects of LAWS in some versions of paragraphs 36 and 39 discussed by the group. The ability to anticipate the effects of a LAWS essentially means predictability.
Reliability is distributed across multiple paragraphs. Paragraph 38(c) seeks to ensure the stability in the performance of the system. Paragraph 40(a) on testing and paragraph 41 regarding biases in datasets also speak to reliability.
Explainability is reflected more broadly in paragraph 40, where the chapeau refers to the need for the expected use and anticipated effects of LAWS to be “understood”. Paragraph 45 seeks to ensure the understandability of the capabilities and limitations of LAWS.
Paragraph 47
Traceability is perhaps the concept least represented in the document. Paragraph 47 deals with effective internal mechanisms for investigation, reporting and appropriate action. This is where traceability would have the greatest relevance operationally.
Traceability could be incorporated by adding a sentence to paragraph 47. This could read as follows:
States should retain records, such as operational logs, sufficient to enable them to recount and review the functioning of LAWS.
With this amendment, even if consensus on paragraph 40(c) could not be reached, its substance would be preserved in other parts of the document.
I thank you, Chair!
