JMIR Publications

ABSTRACT

The emergence of artificial intelligence (AI), particularly large language models (LLMs), has generated interest in their potential role in forensic psychiatry report writing. Although these systems demonstrate strong performance on knowledge-based tasks and can assist with documentation, summarization, and data extraction, forensic reports differ fundamentally from routine clinical notes. They are expert opinions prepared for legal contexts and must demonstrate objectivity, a robust factual foundation, explicit reasoning, and a clear link between evidence and psycholegal conclusions. This viewpoint examines whether AI can and should be used in this context, drawing on current empirical evidence and emerging legal and ethical guidance. Despite promising technical capabilities, LLM outputs remain vulnerable to omission errors, hallucinations, and bias. These limitations are particularly consequential in forensic settings, where small inaccuracies may alter the interpretation of risk, credibility, or responsibility. In addition, AI systems lack intrinsic accountability and may obscure the reasoning process required for expert evidence. Across professional bodies, courts, and regulatory frameworks, there is consistent emphasis on nondelegable expert responsibility, transparency, and the need for human verification. Confidentiality and data governance concerns further limit the use of public AI tools for identifiable case material. A defensible approach distinguishes between low-risk and high-risk uses. AI may be appropriate for bounded support tasks, such as document organization and administrative drafting within secure and validated environments. However, its use for generating or materially shaping substantive analysis or final opinions is not supported. AI can therefore be used in forensic psychiatry report writing only in a constrained and supervised manner, where the expert retains full authorship, accountability, and justification of all conclusions.