Our approach to AI safety

1. Patient data stays on the clinic PC.
2. An unmeasured value stays empty. The software does not fill one in.
3. The doctor reviews every value. A report stays a draft until a doctor sign-off is stored.
4. Dose numbers and the trigger-size status come only from the rule files, and they are suggestions.
5. The software does not diagnose, prescribe, or choose a trigger.
6. A trained model reaches a clinic only through the model release policy.
7. Nothing here is a claim of clinical performance, a regulatory status, or a doctor's approval.

These practices follow the principles in the ICMR Ethical Guidelines for Application of Artificial Intelligence in Biomedical Research and Healthcare (2023).

1. Autonomy
2. Safety and Risk Minimization
3. Trustworthiness
4. Data Privacy
5. Accountability and Liability
6. Optimization of Data Quality
7. Accessibility, Equity and Inclusiveness
8. Collaboration
9. Non-Discrimination and Fairness Principles
10. Validity

The names above are the section titles in that document. ICMR, 2023