International Law and the Regulation of Autonomous Military Capabilities

For nine years, a Group of Governmental Experts (GGE) has been discussing the regulation of ‘emerging technologies in the area of lethal autonomous weapons systems (LAWS)’.¹ (Instead of this rather unwieldy formulation, in this article I will use the term autonomous military capabilities [AMCs] to emphasize the breadth of military applications of autonomous technologies beyond the autonomous weapon systems that the GGE is focusing on.) These discussions have yielded consensus on, inter alia, the subjection of AMCs to international humanitarian law (IHL) and human control and responsibility. Yet it remains unclear how IHL might apply to AMCs and how human control and responsibility might be maintained. Indeed, even defining AMCs is proving difficult.

The dissonance between the GGE’s principled consensus and operational dissensus provides the premise for this article. I will explore how the regulatory framework for AMCs has emerged in the face of this dissonance. By understanding the mechanics of this emergence, we may understand the conditions of possibility and development of this regulatory framework. In turn, this inquiry may shed light on international regulatory frameworks more broadly, especially in relation to new technologies.

I will start, in section 2, by outlining the emerging regulatory framework for AMCs and then defending, in section 3, its regulatory significance. Dissonance between principled consensus and operational dissensus and the GGE’s failure to produce a legally binding outcome call into question the existence of a regulatory regime, emerging or otherwise. I will suggest looking past the superficial uncertainties of this framework to recognize that important regulatory decisions have already been made and that potential regulatory outcomes have been foreclosed. Prohibition or restrictive regulation has been excluded, and the possibility of regulation has been assumed, even though the details of that regulation – IHL compliance and human control and responsibility – remain uncertain. The mechanics of this exclusion (of alternative regulatory possibilities) and assumption (of the possibility of regulation) demand closer scrutiny.

In section 4, I will consider the exclusion of alternative regulatory possibilities. I will restrict my inquiry to one set of alternative outcomes – prohibition or restrictive regulation, which is presumptively impossible on account of the military utility of AMCs. I will unpack this presumption using the example of the everywhere-forever war on terror: the failure of politics that produces the war on terror secures its degeneration into the everywhere-forever war and renders the omniscient infallibility of AMCs presumptively irresistible. Building on this example, I will suggest that this failure of politics is not restricted to armed non-state actors and extends to a broader, amorphous and expanding ideal of security. Within a rationality that defines threats expansively and fetishizes their elimination, the post/super-human promise of AMCs to predict, intercept and prevent threats cannot be ignored, and prohibition or restrictive regulation is implausible.

If prohibition is impossible, we default to regulation, but the impossibility of prohibition does not substantiate the possibility of regulation. When the technological landscape remains uncertain and the principled agreement remains operationally hollow, the possibility of regulation is far from certain. Yet this possibility has been assumed. How? This is the question I will take up in section 5, where I will outline four discursive strategies that sustain the assumption that AMCs are amenable to regulation. Through conflation, different types of AMCs are subsumed within a single imaginary that is more easily accommodated within the emerging regulatory framework. Deferral creates a façade of consensus while shifting contentious issues from the international to the national sphere, where their piecemeal resolution creates vested interests that constrain the possibilities of international regulation. Normalization operates to de-emphasize the novelty of AMCs, while valorization pulls in the opposite direction by exaggerating their virtues. Conflation and deferral obscure regulatory challenges, while normalization and valorization produce a sympathetic regulatory object. Together, these strategies grind down the awkward edges of both the regulatory object and framework to produce an interlocking fit.

The study that I have sketched above is evidently limited in its scope. I will not set out a comprehensive account of the GGE’s regulatory framework. In exploring the exclusion of alternative regulatory possibilities, I will restrict myself to a vignette on the improbability of prohibition or restrictive regulation. In explaining the assumed possibility of regulating AMCs, I will focus on discursive strategies to the exclusion of the agency which animates them and my exposition of these strategies will be illustrative rather than exhaustive. The analysis presented here falls short of a critique of the emerging regulatory framework for AMCs. Any such effort would require contextualization against the broader political economy of resort to force² and attention to the agentive actions and interactions that shape the regulatory process.³ A study along these lines is beyond the present scope. My objective in this article is more modest – to excavate and develop some elements of a critique⁴ and, thereby, to gain insight into the conditions of possibility for the regulatory framework for AMCs and for new technologies more broadly.

In this vein, in concluding in section 6, I will sketch the broader implications of the analysis presented here, specifically the concealed necessity (false contingency) of the extant regulatory framework for AMCs. Acknowledging these structural constraints allows for better appreciation of what is entailed in resisting the emerging regulatory framework for AMCs and for shaping international law’s engagement with other new technologies.

In this section, I will set out a brief overview of the GGE’s discussions, beginning with a justification of my focus on the GGE as the site of the international regulation of AMCs.

The GGE has been convened within the framework of the Convention on Certain Conventional Weapons to produce consensus recommendations on the regulation of AMCs for consideration by the convention’s high contracting parties.⁵ After nine years, the most tangible product is a set of 11 guiding principles.⁶ These principles are ambiguous and lack legal effect, but building upon this meagre consensus has proven difficult. Confounding expectations, the 2021 session of the GGE failed to produce a legally binding outcome.⁷ Disagreement on the form of regulatory outcome and underlying substantive issues were used by a small group of states to obstruct consensus.⁸ The same disagreements obstructed the 2022 session.⁹ While many states continued to demand a legally binding instrument,¹⁰ some preferred principles and practices¹¹ and manuals on the application of IHL to AMCs.¹²

The seeming impossibility of a legally binding instrument at the GGE has prompted calls for shifting to a different forum that is unhindered by consensus requirements and negative vetoes.¹³ However, the obstructive states are the ones who are most actively developing AMCs, so the immediate practical utility of this shift is unclear. Thus, for the time being, it is the GGE that remains at the centre of international discussions and the GGE’s framework that I will use as a proxy for the regulation of AMCs.

To outline this framework, I will describe four key elements.¹⁴ My objective in this description is to highlight the uncertainties that undermine the GGE’s efforts, the consequent evolution of the GGE’s approach and the emergence of a principled consensus. Consequently, the focus is on the consensus positions outlined in the GGE’s reports rather than on the diverse positions reflected in its discussions.

The first of these elements concerns the characterization of the object of regulation.

The GGE has struggled to define the technology¹⁵ and with the question of whether a definition is required at all.¹⁶ These difficulties arise, in part, from technicalities such as whether existing technologies are included in the GGE’s mandate¹⁷ and whether the mandate is restricted to ‘lethal’ weapon systems.¹⁸ Concerns that the regulation of AMCs should not hinder civilian applications of autonomous technologies may have made the question of definition more significant and difficult.¹⁹

But definitional difficulties also reflect technological uncertainties. To begin with, there are significant and unresolved concerns about the functioning of AMCs – the extent to which they will be compromised by bias²⁰ and poor data quality;²¹ whether their functioning will be explainable, predictable and reliable;²² and whether they will be able to develop common sense²³ – that is, adapt their learning across contexts.²⁴

Then there is the range of technologies encompassed within AMCs. One extreme end of that spectrum is the controversial trope of the killer robot that makes and executes lethal decisions without human involvement.²⁵ At the other end, there are mundane applications of autonomy in military processes for translation, logistics and so on. Between these extremes, autonomous technologies are used broadly in targeting processes to support human actions and decisions²⁶ – to process intelligence,²⁷ identify targets²⁸ or assess collateral damage.²⁹ Acknowledging this diversity has led the GGE to shift its focus from autonomous weapons systems to autonomy as an attribute of weapons systems³⁰ and to replace references to ‘lethal autonomous weapons systems’ with references to ‘emerging technologies in the area of lethal autonomous weapons systems’.³¹

Finally, autonomy – which refers simply to the ability to...