The rapid growth of generative AI, which creates original content, has broadened its uses, including in law and justice. This opening article introduces a series exploring how AI is influencing, disrupting and reshaping criminal justice.
AI divides opinion and invites scrutiny of its limits. The Council of Europe recognizes its unprecedented potential to promote and protect human rights, while warning that it can also undermine human dignity and individual autonomy.
For Civitas Maxima, as an organization working to secure accountability for international crimes, AI raises questions that go to the heart of our work: how evidence is gathered, assessed and trusted. For the victims we represent, the stakes are access to justice, recognition of the harm they have suffered and the protection of their sensitive information.
While remaining cautious and aware of the risks, Civitas Maxima is integrating AI into its work. With due regard for data protection and the confidentiality of those involved, we use AI to identify specific references in documents for further investigation, to assist with translation and to decipher handwriting, while recognizing that it cannot replace a professional translator or an expert assessment. We also use it to transcribe videos, as newer tools can sometimes recognize accents more accurately. For example, we are transcribing selected TRC Liberia interviews to corroborate testimonies and identify potential victims relevant to our Liberian portfolio. In the future, we may also use a locally hosted AI system linked to a case management tool, helping us both to organize our work and to strengthen our analysis.
More broadly, AI can help legal professionals manage case files, process documents, turn paper archives into searchable digital records and restore damaged images, all of which can support open-source investigations. Beyond these applications, some jurisdictions use AI as a deciding tool. For example, COMPAS is an algorithmic tool used in US courts to assess among other, the risk of recidivism of criminal defendants. Its utilisation was challenged before the Wisconsin Supreme Court (see State v. Loomis, July 13, 2016) on the grounds that it could infringe the right to a due process. Nevertheless, the Court upheld its use provided that the assessment does not, on its own, determine guilt or the sentence imposed. These practices raise questions about the automation, or even robotization, of justice, and about the biases it may carry from the data it was trained on.
The diversity of these applications prompts reflection on how to reconcile the use of AI with the inherently human nature of the justice system.
As deepfakes and AI-generated images become more realistic, and more crimes are documented on phones and shared online, the question of what can be trusted as evidence becomes ever more pressing.
The next article will explore how AI affects the burden of proof and reshapes perceptions of evidence.