# AI-interaction psychosis research safe-use matrix v99

| Research theme | Runtime use | Prohibited conversion |
|---|---|---|
| Sycophancy and mirroring | Add amplification-mechanism labels and non-affirming response rules | Do not recreate harmful validating dialogue |
| Hallucinated certainty | Add source provenance and uncertainty prompts | Do not present generated claims as evidence |
| Case reports | Inform broad fictional trajectories and red flags | Do not reproduce allegations as verified facts or identify real people |
| Artificial intimacy | Model attachment, dependency, and human-connection protective factors | Do not have the system claim love, sentience, exclusivity, or need |
| Crisis countermeasures | Add fictional support and crisis-interruption modes | Do not provide medical treatment, diagnose, or promise emergency monitoring |
| Screening frameworks | Structure fictional variables such as sleep, isolation, and usage | Do not score or classify a real visitor |
| Regulation and litigation | Inform disclosure, audit, age, and professional-boundary fields | Do not give legal conclusions or certify compliance |