Towards a Threat Model Framework for Deepfake Harms to Prisoners of War Protections under IHL

The diffusion of Artificial intelligence capabilities has created enormous challenges for international humanitarian law (IHL). In response, a feverish dialogue about how best IHL can respond to the use of AI in armed conflict has grown up, considering primarily AI deployments in autonomous weapons systems, decision-support systems, and during cyberoperations. However, this discourse has often overlooked how synthetic media generation capabilities threaten protections for persons hors de combat and prisoners of war (POWs). Our work represents one effort to begin filling this gap.

In our talk, we describe a new threat model framework for synthetic media violations of prisoner of war protections in IHL (specifically, as provided for in the Third Geneva Conventions). We make three contributions: (1) we reorient discourse from Hague-law concerns (the conduct of hostilities) toward Geneva-law concerns (outrages against individual human dignity); (2) we present MAESTRO4IHL, an adaptation of the MAESTRO threat-modelling framework that assesses the harm surface synthetic media create for POW entitlements; and (3) we lay groundwork for future technical and policy measures—red-teaming exercises and institutional mechanisms that, we believe, can begin the important work of addressing AI vulnerabilities to violative synthetic media generation that harm the most vulnerable in war.

This event is organised by the Department of Digital Humanities as part of its seminar series. The talk will be followed by a wine reception.

About the Speakers:

Wm. Matthew Kennedy

Kennedy is an AI+ Senior Research Fellow at King’s College London whose research focuses on AI evaluations methodology for “hard contexts.” He received his Ph.D. from the University of Sydney in the history of international law and held several prestigious postdoctoral appointments prior to a period in industry, where he took up AI evaluation duties. His empirical work now focuses on developing evaluation and red teaming methodologies to assess AI performance on human rights legal reasoning tasks (he co-leads HumRights-Bench alongside Savannah Thais, Caitlin Kraft Buchman, and Malcolm Langford) and AI for education and knowledge-production domains (e.g. he co-leads L2-Bench alongside James Edgell). Active also in the technical AI governance space, he is or has recently been an affiliate of several institutions and organizations, including the University of Sydney Centre for AI, Trust, and Governance, the Center for AI and Digital Policy (as a research group member), and the Schwartz Reisman Institute at the University of Toronto. From 2024-25, he served as an inaugural ORF US-India AI Fellow. He was a winner (adversarial content) of the 2024 NIST ARIA National Red-Teaming Exercise and co-first author on the operational report presented to federal agency stakeholders.

Nathan Heath

Nathan Heath is an AI safety executive and decision scientist with over 14 years of experience across risk assessment, geopolitics, and AI safety. He currently serves as the Founder & CEO of Syntony, an AI safety startup translating red teaming evaluations into good governance. He is also the Co-Founder of The AIHL Project, a research initiative focused on assessing generative AI threats to protections under international humanitarian law. Nathan consults as a red teamer for numerous frontier labs, specializing in risk areas such as TVE, human rights, humanitarian protection, and information manipulation. He previously spent six years supporting the U.S. Department of Defense in areas such as emerging technology risk, security cooperation, and countering violent extremism, and has also held positions at Harvard Law School, the U.S. Department of State, and the Hudson Institute. His work has been published in various outlets, including War on the Rocks, PRISM, and the Global Network for Extremism and Technology. He received his M.A. in Law and Diplomacy from The Fletcher School.