AI Deepfakes and Victim Impact Statement
Wonkwang University Legal Research Institute · 31.03.2026 · 42 (1) · 83-102
DOI: 10.22397/wlri.2026.42.1.83 ↗Kurz gefasst
Fachartikel in der Zeitschrift Wonkwang University Legal Research Institute, online erschienen am 31.03.2026. Verfasst von Jong Goo Kim.
Abstract (Originalsprache)
This article examines the legal and constitutional issues that arise when artificial intelligence (AI) - especially generative technologies such as deepfakes - is used to deliver Victim Impact Statements (VIS) in criminal proceedings, and it draws implications for the Korean legal system. A recent U.S. sentencing case (the Pelkey case), in which a deceased victim “appeared” in court through an AI-generated digital avatar to address the offender, triggered both acclaim as a victim-centered innovation and serious concerns about procedural fairness and defendants’ rights. VIS can serve legitimate aims by informing decision-makers of the concrete impact of crime and by supporting victims’ participation and emotional closure. Yet, because VIS may influence sentencing through emotional appeal, it has long developed within a tension between victims’ interests and defendants’ constitutional protections. The trajectory of U.S. Supreme Court jurisprudence illustrates this structural conflict: the initial restriction in Booth, the shift toward general admissibility in Payne coupled with an acknowledgment that excessively prejudicial statements may render a proceeding fundamentally unfair, and the controversy surrounding highly emotional memorial videos highlighted by Kelly. Against this background, AI-based VIS has notable benefits: it can “represent” the perspective and voice of a deceased victim, enhance the meaningfulness of the right to be heard, and convey the victim’s individuality and the gravity of harm with heightened vividness. At the same time, however, it raises distinct risks. A deepfake VIS may amount to a constructed narrative rather than an authentic statement by the person concerned; it is difficult to subject an AI avatar to traditional modes of verification such as cross-examination; and its audiovisual realism may amplify undue prejudice. Additional concerns include inequality of arms due to the costs and expertise required to produce sophisticated AI media, and broader authenticity crises in the deepfake era - where parties may exploit uncertainty about what is real and what is fabricated (the “liar’s dividend”), undermining public trust in adjudication. The article then evaluates the Korean framework. Korea’s Constitution guarantees a victim’s right to make statements in criminal proceedings, and the Criminal Procedure Act operationalizes this right largely through the structure of witness examination. Yet AI-mediated statements (deepfake videos, synthetic voice, and related digital media) do not fit neatly within that procedural model, and Korea’s adjudicatory practice - where fact-finding and sentencing are not always sharply separated - may increase the risk that emotionally powerful materials distort judicial conviction formation. The article concludes that rather than adopting a blanket ban, Korea should build procedural “guardrails”: advance notice and pre-screening, full disclosure to the defense with opportunities for objections and rebuttal, clear authentication/verification standards for AI-generated media, and guidance on timing and permissible modes of presentation. Such safeguards are necessary to harmonize the meaningful realization of victims’ participatory rights with the defendant’s right to a fair trial.
Zitieren
APA
Kim, J. G. (2026). AI Deepfakes and Victim Impact Statement. Wonkwang University Legal Research Institute, 42(1), 83-102. https://doi.org/10.22397/wlri.2026.42.1.83
Vancouver
Kim JG. AI Deepfakes and Victim Impact Statement. Wonkwang University Legal Research Institute. 2026;42(1):83-102. doi:10.22397/wlri.2026.42.1.83
BibTeX
@article{Kim2026,
author = {Kim, Jong Goo},
title = {AI Deepfakes and Victim Impact Statement},
journal = {Wonkwang University Legal Research Institute},
year = {2026},
volume = {42},
number = {1},
pages = {83--102},
doi = {10.22397/wlri.2026.42.1.83}
}
EUVIMED