As outlined by the address before in this discussion, the conduct of legal reviews is an existing obligation under Article 36 of [AP I], reaffirmed in our context for Guiding Principle (e). It is the task of our expert group to give additional guidance on how it applies in the case of [AWS] taking into account their very complex features. One of the main concerns and a significant challenge is the necessity of such a weapon to remain explainable, understandable and predictable, which stems from the specificities of weapons based on algorithms that complex datasets. Another crucial question with regard to autonomous weapons is at what point in time new legal reviews are necessary. Here again, self-learning capabilities are a significant problem, where after a certain time of operation, it cannot be considered the same weapon anymore. As underlined by others before, the modification of weapons requires new reviews. And the specific problem in this case would be if modifications could be undertaken by the weapon itself. The ICRC rightfully also points out the question of thresholds in such a modification which would trigger a new review. There is a series of technical challenges in the assessment of these weapons and we agree that international cooperation and information sharing in this regard is highly necessary, as are of course also legal safeguards in the form of an international instrument. We note that US-led paper under Article 24 provides us with very useful elements, as do the working papers by a group of 13 states, the working paper by a number of European states, as well as the new working papers by Pakistan and the State of Palestine. There is a considerable amount of useful language available by now concerning legal reviews and related problems, also building on years of work in previous teaches. All of this should now be included in this year’s report. Referring to the content of our own working paper, we consider the requirement, throughout research, development, acquisition and use of [AWS], to constantly review and reassess any possible changes and modifications in the system’s functioning. This should include technical aspects such as machine learning and any datasets upon which system functions are based. This monitoring process should be embedded in an adequate multi-layered international regulatory framework that would entail regular review of the implementation of {prohibitions} and positive obligations in order to ensure meaningful human control is preserved over the use of force, and legal and moral rules and ethical principles are protected in the design, development and use of [AWS].
Statement by Austria under agenda item 5, topic 5 (9 March 2023, transcript)
