Summary
- Sets ten patient-centric ethical principles: autonomy; safety and risk minimisation; trustworthiness; data privacy; accountability and liability; optimisation of data quality; accessibility, equity and inclusiveness; collaboration; non-discrimination and fairness; and validity.
- Puts humans in charge: AI for health "should always be appropriately supervised", using a human in the loop model. Patients should be told about AI use and may refuse it.
- Gives phase-wise guidance for development, validation and clinical deployment, plus an ethics review procedure for ethics committees (ECs) and an informed consent section.
- Ends with an ethics checklist that research teams fill in (objectives, technology, funding and conflicts, team credentials, participants, risks and more).
- Describes itself as a living document to be updated. ICMR's page was last updated 10 Nov 2025, but no revised edition of the guidelines was found as of 2026-10-19.
Who it applies to
- All AI tools built for biomedical and health research and applications involving human participants or their biological data.
- Named audience: health professionals, technology developers, researchers, entrepreneurs, hospitals, research institutions, sponsors, funders and ethics committees.
- Covers both developers (design, training, validation) and deployers (hospitals, clinicians).
- Not extraterritorial. A foreign vendor is reached only through an Indian research site or hospital that follows the guidelines.
Key dates
- Guidelines released (month from press reports; ICMR gives the year 2023 only)
Full date record
| Date | Event |
|---|---|
| 2023 | Guidelines published by ICMR (year from the document and ICMR web page) |
| March 2023 | Public release reported in the press (secondary; exact day not stated in the source) |
| 10 Nov 2025 | ICMR web page last updated (page date, not a new edition) |
Penalties and enforcement
- No penalties stated.
- Practical enforcement is through ethics committee review: IEC approval is required even for using retrospective data to develop and validate AI.
- The guidelines say legal responsibility for an AI tool should be defined before clinical or public use, and that liability depends on the cause: design flaws point to the developer or manufacturer, faulty implementation points to the user or organisation.
- Data privacy text predates the DPDP Act 2023 (it refers to the IT Act section 43A, the DISHA Bill and the earlier Personal Data Protection Bill). Read it alongside current law.
What it means for an Indian company
- Healthtech and medical AI vendors: treat this as the ethics checklist your hospital and research partners will apply. Prepare answers before an EC asks.
- Human in the loop and patient choice: SC-AI-026 Human oversight and override and SC-AI-022 Disclose AI interaction to users.
- Consent, user control over data, access and deletion: SC-AI-011 Lawful basis and consent for personal data used in AI and SC-AI-012 Data minimisation and retention for AI.
- Data quality and representative training data: SC-AI-010 Training and input data governance (provenance, quality) and SC-AI-013 Data bias and representativeness checks.
- Validity, clinical validation and fairness: SC-AI-015 Pre-deployment testing and validation, SC-AI-016 Independent model validation, SC-AI-017 Fairness and discrimination testing of outputs.
- Accountability and liability split agreed before deployment: SC-AI-002 AI roles, accountability and oversight committee and SC-AI-007 Third-party and vendor AI due diligence (contracts with hospitals).
- Ongoing monitoring and grievance routes: SC-AI-020 Performance monitoring and drift detection and SC-AI-028 Customer grievance redressal for AI.
- Software for a medical purpose can also be a medical device under the Medical Device Rules 2017 (the guidelines flag this). That is a separate, binding regime; check it with regulatory counsel.
Obligations
Full obligation breakdown coming
We break the most requested policies down into numbered obligations mapped to controls. Leave your details and we’ll tell you when ICMR AI in Healthcare Guidelines is done.
Glossary
- ICMR: Indian Council of Medical Research, India's apex body for biomedical research.
- DHR: Department of Health Research, the ministry department ICMR sits under.
- EC / IEC: Ethics Committee / Institutional Ethics Committee, which approves research involving people.
- Human in the loop (HITL): a design where a person supervises and can override the AI.
- SaMD: Software as a Medical Device, software that is itself a medical device.
- Living document: a document meant to be revised over time.
- Retrospective data: existing patient records reused for a new purpose, such as training AI.
Key takeaways
- Voluntary ethics guidance, but it gates research approval through ethics committees.
- Ten principles, with human supervision, informed consent and pre-agreed liability as the hard edges.
- Healthtech vendors should map the ethics checklist to their own evidence pack.
- "Regulatory Guidance" overstates it; ICMR is a research body, not a regulator.
Official sources
- ICMR Ethical Guidelines for Application of AI in Biomedical Research and Healthcare (PDF, 2023), accessed 2026-10-19
- ICMR guideline web page (metadata, ISBN, page count), accessed 2026-10-19
- Secondary, for release month only: The Hitavada, 27 Mar 2023 (search result, not opened)
This is general information, not legal advice. Last verified 19 Oct 2026 against the official source. Laws change; confirm the current position with qualified counsel before acting.