Skip to main content
en
Close

Contemporary accomplices? Considering the potential liability of AI targeting system providers under the Rome Statute

, Accountability / Analysis / Artificial Intelligence and Armed Conflict / Artificial intelligence in military decision-making / Emerging Voices / IHL 11 mins read

Rosa Villar Orero  PhD candidate, University of Aberdeen

Contemporary accomplices? Considering the potential liability of AI targeting system providers under the Rome Statute

Photo credits: Unsplash.

Artificial intelligence (AI)-enabled targeting systems are increasingly being used in armed conflicts to process data and provide lists of military objectives. This raises a difficult question that has received little attention: Could the senior executives of the technology companies that provide these systems face criminal liability under international law when their systems contribute to serious violations of international humanitarian law?

In this post, part of our Emerging Voices series, Rosa Villar Orero, PhD candidate at the University of Aberdeen, argues that Article 25(3)(d)(ii) of the Rome Statute, as operationalized through the Katanga test, offers a potential doctrinal avenue for establishing the criminal liability of AI targeting system providers whose contribution is made with knowledge of a group’s intention to commit such crimes.

ICRC Humanitarian Law & Policy Blog · Contemporary accomplices? Considering the potential liability of AI targeting system providers under the Rome StatuteIn 1946, a British military tribunal convicted Bruno Tesch for complicity in the murder of millions of people at Auschwitz. Tesch was a businessman who ran Tesch & Stabenow, a company that supplied concentration camps with a poisonous gas called Zyklon B. Although he never entered a camp and never issued an order to kill, merely sat in his office, processed contracts, and shipped chemicals, he was nonetheless found liable.

In 1946, a British military tribunal convicted Bruno Tesch for complicity in the murder of millions of people at Auschwitz. Tesch was a businessman who ran Tesch & Stabenow, a company that supplied concentration camps with a poisonous gas called Zyklon B. Although he never entered a camp and never issued an order to kill, merely sat in his office, processed contracts, and shipped chemicals, he was nonetheless found liable.  The reasoning was simple: Tesch had knowingly provided the means by which others committed mass atrocity. Physical proximity to the crime was not relevant; what mattered was the structural relationship between his conduct and the crime.

Nearly eighty years later, the question that emerges is: what would Tesch look like today? He would not be selling chemicals. Instead, would he be selling algorithms? As AI services are increasingly used by parties to armed conflict in the conduct of hostilities, it is time to take a fresh look at the precedent for holding companies that provide products and services to warring parties accountable for their contribution, in the case that those parties commit serious violations of international humanitarian law.

The contemporary problem

In contemporary armed conflicts, AI-enabled targeting systems are being deployed to process vast amounts of data to generate lists of military objectives. These systems dramatically increase the speed and scale of target production. They correlate data, find patterns, and recommend targets – in some reported cases, at a rate that human analysts could never match.

Behind these systems stand technology corporations. Their executives are not in the field or issuing strike orders. They are sitting in urban centers often far from the physical site of attacks, approving contracts. Yet, if these systems are deployed in the commission of war crimes at a large scale, the support those technologies provide becomes legally significant, bringing the following question:

Can the senior executives of technology companies whose AI systems contribute to the commission of war crimes be held criminally liable under the Rome Statute?

The doctrinal gap

The immediate answer that international criminal law offers is Article 25(3)(c) of the Rome Statute: aiding and abetting. But there is a critical problem. Article 25(3)(c) requires that the contributor act for the purpose of facilitating the crime.

Technology corporations are primarily driven by the maximization of profits. Their AI systems are designed and sold as legitimate commercial products. Accordingly, proving that a CEO approved a contract with the purpose of facilitating a war crime is, in practice, evidentially difficult.

The consequence of such a heightened standard is a dangerous gap. Corporations that provide the infrastructure of modern atrocity may escape liability simply by arguing that their business was driven by commercial indifference rather than criminal purpose. This problem, however, is not new. Academia has long grappled with these neutral contributions trying to find an answer for the main question: when do neutral contributions become complicity?

The operative avenue

Considering this bottleneck, this post identifies a different provision as the operative avenue. Article 25(3)(d)(ii) provides that a person is criminally responsible if: “In any other way contributes to the commission or attempted commission of such a crime by a group of persons acting with a common purpose. Such contribution shall be intentional and shall either: (i) Be made with the aim of furthering the criminal activity or criminal purpose of the group, where such activity or purpose involves the commission of a crime within the jurisdiction of the Court; or (ii) Be made in the knowledge of the intention of the group to commit the crime.” This is a residual mode of accessorial liability that has received comparatively little scholarly attention.

Unlike subparagraph (c), this provision does not require purpose. It requires knowledge – specifically, that the contribution was made with knowledge of the group’s intention to commit the crime.

This is not a radical departure. It is, in fact, the return to the standard applied by the tribunals that convicted Tesch. Moreover, this provision was operationalized by Trial Chamber II in Katanga. Katanga arose from the ICC’s prosecution of Germain Katanga, a Congolese militia commander, for war crimes and a crime against humanity committed during the 2003 attack on Bogoro, a village in the Ituri district of the Democratic Republic of Congo. In its 2014 judgment, Trial Chamber II shifted his mode of liability from principal perpetration to accessorial liability under Article 25(3)(d). This was the first instance in which the provision was substantively interpreted and in doing so, the Chamber developed a five-element test.

This post argues that this test, properly understood and extended, reaches the conduct of senior executives of technology corporations. However, before applying it, doctrinal foundations need to be established.

Beyond the corporate veil

AI systems are provided by corporations, but the Rome Statute only holds natural persons liable under Article 25(1). Corporate criminal liability was deliberately excluded from the Statute.

However, the exclusion was not a denial of the role played by companies; it was a design choice that requires identifying the right individual within the corporation. The natural person who held effective decision-making power – the executive who authorized or had the power to veto the provision of the system. The I.G. Farben standard, developed after the Second World War, provides the template: a defendant is liable if they were a participant in the act, or, being aware of it, had the authority to prevent it and failed to do so.

The Zyklon B tribunal inferred Tesch’s knowledge from his professional position and the public context of the atrocities being committed. However, the Krupp case refined this and established that it cannot be inferred from the defendant’s official position alone, but it must be proved that the contributor “knowingly participated”.

Accordingly, the knowledge-based mens rea proposed in this post does not emerge in a vacuum; it was developed through the jurisprudence of post-WWII military tribunals, later consolidated by the ad hoc tribunals, and now forms part of customary international law.

Katanga test applied

With those foundations in place, the five elements of the Katanga test can be applied to AI-enabled targeting systems, and grounded in contemporary conflicts.

Element (i): A crime within the jurisdiction of the Court was committed. The ICC has issued arrest warrants in contemporary armed conflicts for the war crime of intentionally attacking civilians under Article 8(2)(b).

Element (ii): The persons who committed the crime belonged to a group acting with a common purpose. Trial Chamber II in Katanga argued that the existence of such group “would be evidenced without necessarily establishing the existence of an organization incorporated into a military, political or administrative structure”. It may conversely be inferred that a formal military structure is itself strong evidence of organized group.  The common purpose or shared intent of the group is to be established under Article 30(2)(b) and inferred by the targeting patterns of armed forces.

Element (iii): The accused must have contributed to the crime. This element is the most doctrinally contested. Departing from a strict but-for causation test, this post applies a risk-based approach. The AI system need not be indispensable to the crime, but its provision must have materially increased the risk of its commission.

AI targeting systems process data and generate targets at unprecedented speed and scale, considering the error rates and loose collateral damage thresholds; this directly has a bearing on the occurrence and the systemic manner in which the crimes occur.

Element (iv): The contribution must be intentional. Intentionality refers to the act of providing the system, not to the criminal outcome. In this regard, commercial transactions are by definition deliberate and contract-driven. Accordingly, this element is the most readily demonstrable.

Element (v): Knowledge of the intentions of the group. The AI systems providers’ knowledge of the intention of the group, must be defined with reference to Article 30(3) of the Statute. Under this Article, the provider must have been aware of the group’s intention to commit crimes at the moment of providing the system. Here, a contested question arises: does that awareness need to be established for each specific crime, as Katanga requires, or does general knowledge of the group’s criminal intentions suffice, as Mbarushimana holds? This post aligns with Mbarushimana since requiring specific knowledge per crime would implicitly give the provider control over each specific attack, and that level of control would expose them to a stronger mode of liability than Article 25(3)(d)(ii).

Proving such awareness in cases where merely punctual war crimes are committed may be an arduous task, since war crimes are likely to be committed in any armed conflict. However, in contemporary conflicts, persistent atrocities are rarely unreported. NGO reports, journalistic investigations, and ICC proceedings generate publicly available documentation. It is that documentation which is the proper lens through which to assess whether the provider could have expected their system to be used exclusively for lawful purposes. In cases where significant civilian harm is widely reported as resulting from military operations in which technologies play a central supporting or enabling role, a corporate executive with decision-making power over relevant transactions could struggle to credibly claim ignorance of the general pattern of violence.

Conclusion

In 1946, Bruno Tesch was convicted without ever setting foot in Auschwitz. He was found guilty by knowingly providing the means by which others committed mass atrocities. In 2026, the question that arises is how that principle can be applied to accountability in relation to contemporary atrocities, specifically, in the event that technology companies and executives provide algorithms that are central to the commission of serious IHL violations.

This post contends that the relevant legal architecture already exists in the Rome Statute. Article 25(3)(d)(ii), as operationalized through the Katanga test, offers a proper doctrinal pathway that rather than an overextension of the Statute, is a way of reimagining accessorial liability for the algorithmic age.

In an era when international law is facing unprecedented skepticism and is perceived as failing to adequately address the commission of atrocity crimes, the question is whether it seeks to shield those who tacitly approve of international crimes, or whether it is willing to assume responsibility and make use of the available legal mechanisms through a dynamic interpretation tailored to contemporary challenges.

 

Related articles

Share this article
You should also read these articles