Civilian Protection in the Age of Deepfakes

[Flóra Merész is a Public International LL.M. graduate from Leiden University, a former intern in the ‘Regulation in the public interest: Disruptive technologies in peace and security’ strand of the T.M.C. Asser Institute, currently working as a Legal Fellow at Just Access e.V.]

Deepfakes not only blur reality in our everyday life but also increasingly on the battlefield. From fake videos of Zelensky announcing Ukraine’s capitulation to AI-generated clips of attacks that never happened in the Iran war, deepfakes are used as a new tool of disinformation without much constraint. They not only create chaos and distrust among combatants and the public but also have the potential to influence the ability of civilians to make sound and informed decisions. The proliferation of deepfakes in armed conflict invites reflection on the current civilian protection frameworks under international law regarding the permissibility of disinformation.

Information as a “Survival Right”: Deepfakes and the Threat to Civilian Safety

Access to accurate and trustworthy information during armed conflict is vital for ensuring the safety of civilians. In fact, the right to information has been termed a “survival right” by the Special Rapporteur on freedom of expression and opinion. Civilians rely on information from various sources, including increasingly from social media, to determine where to flee, where safe corridors are, where to seek care, how to secure fundamental necessities or whom to trust.

The deployment of information-related capabilities with the aim of influencing or manipulating the behaviour and attitude of the adversary or civilian populations to achieve strategic or military goals has always been a part of armed conflict. What has changed is the speed, scale and reach of such activities thanks to technological developments that enable the immediate creation and transmission of information across geographical boundaries at a low cost. This lowering of technical barriers is especially worrying in the case of disinformation, defined as “the manipulation of false or misleading information to intentionally deceive and cause public harm” (emphasis added). While historically directed at enemy soldiers, today, disinformation operations increasingly target civilians. 

Deepfakes, a form of synthetic media, including images, videos or audio, generated or modified by AI to appear authentic, have emerged as a powerful tool of disinformation. Unlike conventional methods of disinformation, deepfakes are “convincing, scalable, and increasingly accessible”.  Today, thanks to the emergence of Generative AI, anyone can create hyper-realistic fake content in a matter of seconds and circulate it on social media with very few barriers. Deepfakes have become so sophisticated that even leading digital forensics experts struggle to tell apart false content from reality. For civilians, who largely rely on information to ensure their safety amid hostilities, the manipulation of believable information at the speed and scale enabled by deepfakes may have fatal consequences. 

For AI-generated synthetic content, the sky is the limit, mimicking figures in positions of trust and authority, such as military or political leaders, journalists, or humanitarian organisations, and spreading false instructions or information. Deepfakes can depict convincing imagery of ongoing hostilities, attacks targeting essential services, refugee camps or even nuclear sites. Consider the following illustrative examples, inspired by recent events:

• a deepfake video of a local militia leader instructing civilians to evacuate through a certain safe corridor, where, in fact, there are active hostilities;
• a deepfake video or pictures of an overwhelmed or attacked hospital, discouraging civilians from seeking essential medical care;
• a deepfake video of a journalist reporting about desperate conditions in civilian neighbourhoods or refugee camps, such as the lack of electricity, clean water, or food, redirecting civilians who may have planned to flee there;
• deepfakes of humanitarian organisations claiming or showing that they have no more food or medical supplies, leaving civilians in precarious conditions.

What is common in these scenarios is that they might fall through the cracks of international humanitarian law (IHL), leaving civilians without adequate protection.

The Limited Reach of IHL

While technology-neutral (and thus equally applicable to deepfakes), IHL takes a “remarkably lenient approach” to disinformation. This is due in part to its traditional focus on the kinetic means and effects of war and to the historically limited reach of information operations, primarily affecting combatants. 

The dissemination of deepfakes in disinformation operations aimed at civilians raises questions about its characterisation under IHL. Disinformation has been traditionally considered a permissible ruse of war. Although IHL adopts a surprisingly permissive stance to ruses, this does not render their use wholly unconstrained.  

First, permissible ruses are those that are aimed at achieving a certain military objective. Article 37 AP I defines ruses as “acts which are intended to mislead an adversary or to induce him to act recklessly”. In other words, ruses of war are measures primarily aimed at placing the adversary (generally understood to refer to enemy soldiers or fighters, not civilians) into a disadvantageous situation, and as a result, securing a military advantage. The 1987 Commentary’s definition of ruses as “inducing an adversary to make a mistake by deliberately deceiving him, or … inducing him to commit an imprudent act” leads to a similar conclusion (para. 1515). Understanding ruses under IHL as actions primarily targeting enemy forces, rather than civilians, has also been advanced in scholarship (see here, here and here). This is an important qualification as today, the manipulation of the civilian information space during an armed conflict may have the sole aim of spreading confusion among civilians without necessarily having a nexus to hostilities.  

IHL further limits permissible ruses by prohibiting perfidious acts and, more importantly for our context, by conditioning their lawful use on compliance with international law applicable during armed conflict. As will be explored, international human rights law (IHRL) obligations applicable in armed conflict may place important restrictions on the use of deepfakes.

Additionally, it is worth exploring whether deepfakes in this context can, more generally, be classified as “military operations” or fall into the even narrower category of “attacks”. Such characterisation has the benefit that, at a minimum, Article 57(1) AP I would be triggered, obliging parties to exercise “constant care” to protect civilians and civilian objects. Further, if the dissemination of deepfakes amounts to an “attack”, the principles of distinction, proportionality and precautions would all apply.

While historically military operations have been understood to refer to physical conduct, it has been argued that information operations intended to further military objectives could fall into the category of military operations, making them subject to the obligation to take constant care. Although somewhat abstract, it is generally accepted that it contains a self-standing binding obligation requiring continuous situational awareness and the consideration of the effects of activities on the civilian population. As a result, when deepfakes have a nexus to hostilities, their effects on the civilian population must be considered.

The scope of “attacks”, understood as acts of violence against the enemy, is narrower. It is now widely accepted that it refers not only to violent means (i.e. kinetic force) but also to violent consequences. As a result, deepfakes or other information operations that result in death or injury may qualify as attacks. Notably, however, the nexus between the harm and the operation must be sufficiently direct, which would be difficult to prove in the case of deepfakes. 

What then are the limits to using deepfakes against civilians in armed conflict?

A Patchwork of Protections Offered by IHL and IHRL

Although deepfakes are difficult to classify within the existing framework of IHL and therefore remain largely unregulated, the law of armed conflict does provide several fallback options for addressing their use. Additionally, IHRL equally applies in times of armed conflict, providing additional layers of protection. As acknowledged by the International Court of Justice, IHRL may not only complement issues also regulated by IHL but also address questions that are not regulated by this field of law at all or where rules remain rudimentary, such as the right to freedom of expression or the right to health.

Protections Afforded by IHL

Firstly, IHL offers heightened protection to certain objects and services vital for securing human needs. Obligations to respect and protect medical units and humanitarian operations may act as a limit to the lawful use of deepfakes. Additionally, humanitarian personnel and operations must not be harmed in any way (Rules 31 and 32 ICRC Customary IHL Study). The dissemination of deepfakes with the effect of breaking trust in humanitarian organisations or discouraging civilians from seeking humanitarian aid in any other way would run counter to States’ obligations under IHL. A similar logic can also be applied to medical units. Reflecting customary international law, multiple military manuals prohibit any interference with medical units that would prevent them from performing their medical duties. Spreading false information about hospitals or clinics by means of deepfakes would thus be unlawful.

Furthermore, it has been argued that disinformation that dissuades civilians from obtaining food, water, or other “objects indispensable to the survival of the civilian population” by prompting unwarranted doubts about their safety could be seen as “render[ing] useless” those objects, thus violating Article 54(2) AP I. Indeed, the Commentary explains that the words “attack”, “destroy”, “remove” and “render useless” were chosen to cover all possibilities (para. 2101). Although the ICRC’s explanatory notes state that rendering objects useless “refers mainly to irrigation works and installations”, as the wording indicates, the examples given are not exhaustive (para. 2100). What matters is that the act is carried out “for the specific purpose of denying them [objects indispensable] for their sustenance value to the civilian population”. For example, a video falsely showing a freshwater source as contaminated may breach the provision if it deters civilians from using the source and deprives them of water needed for survival.

Lastly, some also consider that the principles of military necessity and humanity create self-standing obligations, rather than solely informing the interpretation of other rules. In other words, for an operation to be lawful, it must comply with the two most fundamental principles of IHL, even if there is no specific IHL rule prohibiting it. Consequently, the use of deepfakes that is not necessary to achieve the only legitimate purpose of a conflict, “to weaken the military forces of the enemy”, would be unlawful under the principle of military necessity. As a corollary, if the dissemination of deepfakes would be reasonably expected to cause suffering, injury or other harm to civilians, it would run counter to the demands of humanity, rendering it unlawful. 

Human Rights Law Safeguards

Few would contest today that IHRL applies not only in peacetime but also during armed conflict. While its application is often hindered by limits on its extraterritorial application, there are several provisions that are relevant for deepfakes without having to dive too deep into the debates surrounding their territorial scope.

First and foremost, IHL contains no rules related to access to information despite its vital role in ensuring civilians’ safety. This gap is easily filled by IHRL, which guarantees the right to freedom of expression and opinion, including the “freedom to seek, receive and impart information and ideas of all kinds”. Access to “diverse, verifiable sources of information” is a fundamental human right. Consequently, at a bare minimum, States are required to exercise due diligence to ensure the free flow of trustworthy public information, including through independent and pluralistic media on their territory. Further, States may also be under a negative obligation to respect the right to information in other States. This interpretation would be consistent with an expansive and factual approach to the extraterritorial application of human rights.

In addition to the right to freedom of expression, there is a range of socio-economic rights that can provide protection against deepfakes not reaching the threshold of attack but causing significant harm by depriving the civilian population of essential services or objects. The International Covenant on Economic, Social and Cultural Rights contains no clause on its territorial scope. Accordingly, the Committee on Economic, Social and Cultural Rights (ICESCR) has affirmed that the right to health under Article 12 has an extraterritorial dimension (para. 39). Consequently, States must also respect the right to health outside of their territory and prevent third parties from violating such right abroad. By analogy, all other rights protected under the ICESCR, such as the right to adequate food and housing, could be understood as carrying a similar obligation. Failing that, General Comment No. 14 nonetheless clarifies that the right to health encompasses access to safe food, adequate nutrition, and housing as underlying determinants of health. Consequently, deepfakes that undermine access to basic necessities would constitute a violation of Article 12 ICESCR.

Concluding Remarks

Disinformation in armed conflict is a centuries-old practice. However, parties to a conflict have never been able to spread false information at such speed and scale, and in such a convincing manner, as they are today. The growing practice of targeting civilians by means of information and communication technologies is no less alarming than their targeting by kinetic force. The harms caused by deepfakes and other means of disinformation are often indirect and, as a result, invisible. However, the importance of reliable information should not be underestimated. If deepfakes undermine access to and trust in reliable information, civilians lose the single most important self-protection tool, exposing them to immense danger. IHL and IHRL should not be unduly constrained by traditional interpretations and must instead adapt to today’s realities to protect those most vulnerable.