This Arkangel AI whitepaper is a narrative community evidence brief, not a peer-reviewed study, systematic review or clinical guideline, on supportive AI in mental health in Latin America. It synthesizes 26 sources from PubMed/MEDLINE, PMC, Crossref, WHO, the American Psychiatric Association and the FDA (cutoff June 15, 2026), including 5 direct studies from Brazil, Argentina, Colombia and Chile, and grades clinical-text analysis, assisted screening, conversational support, risk prediction, generative documentation and crisis detection by evidence stage.
Evidence is relatively more developed for natural-language processing, assisted screening and structured chatbots: a meta-analysis of 18 randomized trials with 3,477 participants favored short-course chatbots for depression (Hedges g −0.26) and anxiety (−0.19), with no substantial effect at three months. In Colombia, a retrospective study of 13,607 records predicted conversion from depression to bipolar disorder with 72% sensitivity but 38% precision, a signal rather than a decision rule. The authors find no support for autonomous diagnosis, unsupervised generative therapy, individual suicide prediction or automated crisis management, and recommend starting with reviewable documentation, search and prioritization under human accountability.
