By Smith, J., Camello, M., & Planty, M
Generative artificial intelligence (AI) refers to AI1 used to create content, such as text, images, music, audio, and videos.2 Generative AI offers many potential benefits, enabling users to automate, augment, and accelerate a wide range of workflows, from simple administrative tasks like transcription and translation to more-complex functions such as investigation and decision support. In the criminal justice system, generative AI offers promising solutions to address human resource and budget challenges, allowing practitioners to focus on more-impactful work. Generative AI–integrated tools may enhance data analysis, improve detection and objective assessment of evidence, and streamline administrative processes. However, its integration, particularly in the criminal justice domain, raises some concerns, including potential biases, privacy issues, and the need for rigorous oversight to ensure effective implementation. It is unclear whether these tools can deliver on their promised efficiencies in practice, as evidenced by early research evaluating time savings of implementing AI-assisted report writing software.3 These concerns highlight the necessity for addressing bias and accuracy, maintaining strict data privacy and security protocols, and promoting transparency and accountability in AI-driven decisions and processes. This report is intended to help criminal justice decision-makers do the following: ¡ Understand what generative AI is and how it relates to the criminal justice system ¡ Identify how generative AI may be applied to tasks and jobs within the criminal justice system and the potential benefits, realities, and limitations ¡ Consider the technical, operational, and governance factors that may influence adoption and implementation ¡ Understand what makes up the generative AI technology stack and how models can be trained Key Takeaways • Generative AI represents an acceleration and advancement in technological innovation that already impacts the criminal justice system and will continue to do so—it is no longer a question of if or when, but how and to what extent. • Generative AI–powered software tools may offer many potential benefits, such as improving efficiency and augmenting capabilities across an extremely broad set of applications for criminal justice system stakeholders. Although these products hold promise, little empirical evidence currently supports or refutes promised benefits from these products. • Generative AI models can be deployed in various forms, including cloud-based models that centralize data processing and federated models that enable decentralized training across multiple locations, preserving data privacy and enhancing security for sensitive criminal justice applications. • Decision-makers should be aware of the substantial technical, operational, and governance risks associated with generative AI– powered software tools prior to implementation. • Responsible use of generative AI requires addressing bias and accuracy concerns, maintaining strict data privacy and security protocols, adhering to ethics and legal standards, and promoting transparency and accountability in AI-driven decisions and processes. • Generative AI technology is evolving faster than the legal or policy environment for AI—the criminal justice community must be proactive and must implement robust internal training and policy frameworks rather than relying solely on external legal or regulatory guidance.
Research Triangle Park, NC: RTI International, 2025 28p.
