JMIR Publications

ABSTRACT

Background:

Artificial intelligence (AI)-themed delusions are increasingly reported within psychotic-spectrum disorders. While contemporary sociotechnical content is known to shape delusional narratives, it remains unclear whether AI occupying a structurally central role within a delusional system is associated with elevated violence risk or more restrictive judicial outcomes in forensic psychiatry. Given the emphasis placed on insight, adherence, and dynamic risk factors in review board decision-making, understanding whether AI centrality constitutes an independent risk marker is clinically and legally relevant.

Objective:

To examine whether centrality of AI within psychotic delusional systems is associated with (1) violence toward others and (2) judicial findings of significant public safety risk and restrictive dispositions in forensic psychiatric judgments.

Methods:

We conducted a retrospective jurisprudential study of publicly available decisions from Quebec psychiatric tribunals and courts identified through systematic searches of the Société québécoise d’information juridique (SOQUIJ) database. Judgments were included if they involved a psychotic-spectrum disorder and explicit AI-related delusional content. AI centrality was coded as central versus non-central using a structured framework. The primary outcome was documented violence toward others. Secondary outcomes included direct AI-violence attribution and judicial findings of significant public safety risk. Associations were analyzed using Fisher exact tests and odds ratios (ORs) with 95% confidence intervals (CIs).

Results:

Twenty-nine judgments met inclusion criteria. AI was coded as central in 15/29 cases (51.7%). Violence toward others was documented in 20/29 cases (69.0%). Violence occurred in 12/15 AI-central cases (80.0%; 95% CI 54.8-93.0) versus 8/14 non-central cases (57.1%; 95% CI 32.6-78.6), yielding an OR of 2.91 (95% CI 0.63-13.45; P=.26). AI centrality was strongly associated with direct AI-violence attribution (9/15 vs 2/14; OR 9.00, 95% CI 1.48-54.6; P=.014). Judicial findings of significant public safety risk were observed in 13/15 AI-central cases (86.7%) versus 9/14 non-central cases (64.3%) (OR 3.60, 95% CI 0.63-20.5; P=.24). Lack of insight was more prevalent in AI-central cases (13/15, 86.7%) compared to non-central cases (8/14, 57.1%; OR 4.89, 95% CI 0.79-30.1; P=.091). Treatment refusal was also more frequent in AI-central cases (60.0% vs 28.6%; OR 3.75, 95% CI 0.74-18.9; P=.142).

Conclusions:

AI centrality within delusional systems was not independently associated with statistically significant increases in violence toward others. However, AI-central cases were more likely to involve direct AI-violence attribution and markers of epistemic vulnerability, including impaired insight and treatment refusal. These findings suggest that AI may function as a structural organizer of delusional meaning-making rather than as a novel criminogenic risk factor. Forensic decision-making should therefore prioritize dynamic clinical risk variables over thematic novelty.