AI Psychosis Tracker logs 126 cases

This is less a consumer product than a sobering research index, and that distinction matters. The site is useful precisely because it makes a messy, undercounted harm pattern legible, but it also inherits the limits of public reporting.

• –Weekly scraping plus manual review gives the project a repeatable methodology, which makes the dataset more credible than a random link dump
• –Splitting cases from academic papers helps separate anecdote, journalism, and more formal research, which should make it easier to cite carefully
• –Because it only captures what is publicly reported, the number is almost certainly a floor, not a full count
• –The tracker is most valuable to safety researchers, journalists, and clinicians looking for pattern recognition rather than causal proof
• –For builders, it’s a reminder that emotionally tuned chat systems can create real downstream risk, not just abstract alignment concerns