I am very grateful to Marko, Ashley, Philippa, and Paul for their thoughtful contributions. I greatly appreciate their thorough engagement with my work. To conclude the symposium, I will react to a few key themes raised by their comments.
Political and legal implications
I opened this symposium by noting that party status is a legal concept to which States attach significant political implications. The commentators have added further facets on the intricate relationship between legal and political considerations concerning party status. Marko reminds us that policy and legal considerations need not match and that therefore the connection of political red lines to the legal notion of party status is tenuous. Paul underscores the political impact of establishing party status. From a practical perspective, he notes that a clear sense of the relevant legal criteria may mitigate risks of misunderstandings between adverse parties. I would add that States could make use of that risk mitigation by making their views on the criteria public. Doing so would also have benefits within the State for managing the domestic implications of party status that should not be overlooked, as Ashley rightly points out.
I also share Ashley’s argument for recognising the potential of co-parties driving each other to comply with international law in armed conflict, and I could not have made this case more forcefully than she has. Yet Marko is right that the success of this argument is ultimately in the hands of governments. Paul sensibly cautions against imposing additional obligations on co-parties. To be clear, however, I do not argue that co-parties have additional obligations going beyond what party status generally entails. Indeed, co-party status is no separate legal status from party status: a co-party has party status, as one among several parties on the same side of a conflict. My argument is that the obligations international law imposes on parties must be interpreted to discern their meaning in the relationship between fellow co-parties. In other words, co-parties must not do something above and beyond what party status requires, they must do precisely that.
Methods I: normative proximity
In terms of interpretive methods, Marko notes that identifying (co-)parties involves implicit normative judgments on proximity between the conduct of states and military harm It is true that the notion of hostilities—as the building block of the conflict relationship between the parties—being defined as acts intended to cause military harm brings about issues of proximity that are not fully articulated in the law. Contrary to what I had hoped at the beginning of my research, general theories of causation proved to be of little help in drawing lines. And systemic reasoning and international practice suggest responses to the proximity question (‘direct connection to military harm’) that remain, well, approximate. More broadly, Marko is right in hinting that there are implicit normative assumptions in my methodological approach. My aim was to understand the concept of party status within the system of international law’s regulation of armed conflict. As I explained in my introductory post, I have relied on systemic reasoning—and then tested the result of this reasoning against the available practice, rather than attempting to build up my account inductively from international practice—to allow for what I believed to be the richest and most coherent account of party status. At the same time, the very notion that international law is, at least to a sufficient extent, a system that allows for such reasoning is an assumption that is widely accepted but still has normative underpinnings. For my account of party status, systemic reasoning imports normative assumptions across the existing system of international law in armed conflict, including assumptions of proximity woven into notions such as hostilities.
Methods II: reading State practice
Ashley raises another important difficulty of interpretive method, namely that of reading State practice in this realm. Specifically, she wonders whether statements by States applying IHL to their first strikes in joining the anti-ISIL coalition tell us much about party status. As a general matter, I share her caution as to the value of those statements. They can only be ‘an implicit indicator’ that the strikes were considered to be part of a multi-party conflict, rather than being separately assessed against the intensity threshold that would need to be met for the existence of a (new) non-international armed conflict (NIAC) (p 118). The reason for this caution is, of course, as Ashley perceptively observes, that States have merely confirmed their view that IHL applied to these strikes. They have not stated that this was so because they had become a party to a pre-existing multi-party NIAC by virtue of their first strikes. How plausibly we can infer the latter conclusion from the former statement depends on how plausible we consider alternative legal explanations for why States considered IHL applicable.
Ashley suggests that ‘some states now seem to accept that IHL does (and should) apply to one-off military strikes against groups taken in individual or collective self-defense, whether or not those strikes lead to a NIAC’. It is possible that States hold this view, but I have not seen much evidence (which may be due to my oversight or a lack of publicity). Ashley provides a link to a 2020 statement by the then General Counsel to the US Department of Defense, in which he considered the lethal US strike against Qassem Soleimani to be governed by IHL. But this case relates to an international armed conflict (IAC) scenario, where there is no intensity threshold that would need to be crossed, unlike in NIACs. Soleimani was a general in the Islamic Revolutionary Guard Corps of Iran, a member of a State organ. If his killing was not already part of an ongoing IAC between the US and Iran (as has been argued), the question in this case was whether the first strike—which in my view suffices to create an IAC—is itself also regulated by IHL (for opposing views on the rules governing this strike, see, eg, here at 64 and here). I am sceptical that there is ground for applying IHL to a strike against State or non-State actors that is not somehow considered part of an armed conflict. That is just not how IHL is structured, at least in the realm of the conduct of hostilities.
Overall, I still believe it to be plausible that States beginning their strikes against ISIL implicitly took account of the fact that they joined in intense partnered operations. Their application of IHL from the outset can best be explained as a collective intensity assessment. However, Ashley rightly stresses that the evidentiary value of practice remains questionable where States do not provide reasons for legal conclusions that can be explained based on different, competing understandings of the law.
Military Artificial Intelligence
Finally, Philippa makes the important point that the rapid development of military Artificial Intelligence (AI), and particularly Decision Support Systems (DSS) poses challenges for identifying co-parties. The virtue of general rules is that they may cover new technological developments. General rules provide crucial guardrails for new technologies when technology-specific regulation proves difficult (see here)—a tendency to which the deadlocked attempts to regulate lethal autonomous weapons are a vivid testimony. I would tentatively argue that the legal criteria for identifying co-parties are sufficiently general to capture the use of military AI DSS.
Attribution of conduct is, of course, a challenge far beyond identifying party status. For international law to remain workable, we must remain able to ascertain that (and when) a State (or other collective entity) has acted. Provided military AI tools indeed merely support human decisions, attributing the human conduct to the State is sufficient. Philippa’s concern is that AI DSS are (not on paper but empirically) increasingly likely to effectively replace actual human judgment. Normatively, I have serious reservations about non-human decisions in the conduct of hostilities. As a matter of international law, however, even if humans merely implement courses of action that have been effectively pre-determined by AI systems, they own these decisions when implementing them in their conduct. If State armed forces, as organs of the State, employ AI DSS and implement decisions suggested by those systems—or even let the systems themselves implement those decisions—this deployment should be considered conduct of the State under Article 4 ARSIWA (see similarly Paola Gaeta regarding autonomous weapons). Even if, for example, a State employs an AI system to detect and intercept a hypersonic missile, without any human involved (for sheer lack of time) in the specific incident, would we not say that the shooting down was done by the State?
Similarly, regarding the direct connection to hostilities criterion and the supply of an AI DSS by one State to another State (or other collective entity), the question is who deploys and operates the system. Supplying a DSS does not directly cause military harm to the adversary; it is only the actual use of the system that does so. Determining who operates the system may not be straightforward in practice, as evidenced by the debates around the level of involvement required by supplying States to operate certain long-range weapon systems in Ukraine. If the assisting State supplies targeting data generated by an AI DSS while continuing to operate the system as part of specific military operations, this is comparable to traditional intelligence sharing. For purposes of identifying parties, there is no substantial difference between providing AI-generated targeting data and supplying targeting data obtained through conventional intelligence methods. Both can be part of the targeting cycle for a specific military operation. By contrast, if the assisted State itself uses the AI DSS to generate targeting data, the scenario is comparable to the provision of any other military tool. In this scenario, the assistance lacks a direct connection to hostilities.
Finally, I agree that coordination among (potential) co-parties may look different from what it used to in the past. Coordination facilitated by the interoperation of the respective States’ AI DSS may, of course, become even more difficult to discern from the outside, thus increasing the practical challenges for identifying co-parties. Nevertheless, it will still be coordination and there is no compelling reason to suggest that this particular means of coordination should be less relevant for identifying co-parties than more traditional means. Philippa may be right, of course, that AI DSS may diminish States’ awareness of their partners’ (and even their own) actual decision-making processes in the conduct of hostilities. The structural challenge we face, as in other areas of law, when specific decisions are automated, is that we may need to attach greater legal relevance to the decision on the deployment of the automation system. There are reasonable questions as to how much legal work those more high-level decisions can do.
More importantly, however, these reflections should also prompt us to question how much decision-making on the conduct of hostilities should be delegated to AI tools and how we can ensure that parties to a conflict do not lose sight of how and what decisions are made in war.
