This Arkangel AI whitepaper is a narrative evidence brief on which artificial intelligence tasks in oncology have useful supporting evidence, which remain emerging, and what should precede their evaluation in Latin America and the Caribbean. The authors searched PubMed, PMC, Crossref, publisher sites and official ASCO and WHO sources, with a main search on June 14, 2026, and assembled 27 sources in six domains: AI-assisted mammography, digital pathology, language models and information support, health record extraction, clinical trial matching, and regional and governance evidence. It is not a systematic review, and maturity ratings are the authors' narrative judgments.
AI-assisted mammography within organized screening programs had the most mature evidence. In the German PRAIM implementation study of 461,818 women, AI-assisted reading detected 6.7 cancers per 1,000 versus 5.7 without AI, an adjusted relative difference of 17.6% with no increase in recall, and the interim analysis of the randomized MASAI trial (80,020 participants) reported 44.3% less reading workload; neither shows a mortality benefit. Digital pathology has task-specific external validations, but 99% of 100 studies in one review had at least one high or unclear risk-of-bias or applicability domain, while language models, record extraction and trial matching remain retrospective or experimental. Direct Latin American evidence was insufficient to establish regional effectiveness. It is a whitepaper, not peer reviewed, and all authors work at Arkangel AI.
