ICMR Ethical Guidelines for Application of Artificial Intelligence in Biomedical Research and Healthcare (2023)

ICMR is the Government of India's apex body for biomedical research and functions under the Department of Health Research in the Ministry of Health and Family Welfare. Its guidelines page lists dozens of research-ethics and clinical guideline sets, and the AI document appears both under the "Ethics" heading and in its own category, "AI in Biomedical Research & Healthcare 2023".²

The guidelines were prepared by the DHR-ICMR Artificial Intelligence Cell in New Delhi and published in 2023 as a 76-page document with the ISBN 978-93-5811-343-3. ICMR's page names Dr N.K. Arora, Executive Director of the INCLEN Trust, as chairperson. It describes the document as the product of expert consultation and as a "living document" that will be updated.¹ MediaNama, which summarised the guidelines on 5 May 2023, reported that they came out in April 2023.³

The motivation set out in the summary is that AI in health depends on people's data, which raises risks of bias, breaches of confidentiality and patient harm, and that AI-specific guidelines can help assign accountability for errors.³

According to ICMR's description, the guidelines promote ethical conduct in AI for biomedical research and healthcare and provide a framework for ethical decision-making across the development, deployment and adoption of AI-based medical solutions. They cover ethical principles, guiding principles for stakeholders, the ethics review process, AI governance and informed consent. The intended audience is developers, researchers, clinicians, ethics committees, institutions, sponsors and funders.¹

MediaNama's summary describes two parts: a set of ethical principles, and guiding principles for three stages of an AI tool's life, namely development, validation and deployment. The guidelines apply to AI used in biomedical research and in healthcare delivery that relies on health data.³

MediaNama's summary lists the principles as follows (paraphrased):³

The summary does not set out a count or ordering that should be treated as authoritative for the full document. Readers who need the exact wording of each principle should consult ICMR's text.¹²

The stage-specific guidance, as summarised by MediaNama, includes the following:³

• Development. Institutions must disclose how development data will be used, and the data must not cause harm or discrimination. Users and physicians should have a feedback mechanism, and patients who opt out should be able to have their data removed or modified.
• Validation. A multi-sectoral team, including data science, public health and engineering experts, must confirm that the technology is a valid solution for the clinical condition. Validation should take account of user-experience results and risks to both providers and patients.
• Deployment. Health professionals must understand how the technology works and its limitations. Cost-effectiveness and operating costs should be disclosed where available, and patients or their legal representatives must be told the extent to which AI was used in research or clinical decisions.

Two features connect the guidelines to the wider governance of AI. The first is the role of ethics committees: they review AI research proposals, receive reports of identified risks and are expected to check for harms at every stage of a study. The second is the allocation of liability, which the guidelines say must be settled before an AI tool is used rather than after harm has occurred.³

The guidelines are ethical guidance and do not create enforceable duties. They refer to DISHA and the PDP guidelines in the context of international collaborations; the Digital Personal Data Protection Act, 2023 was enacted after the guidelines were published. The sources reviewed for this entry do not describe how the guidelines are enforced, for example by institutional ethics committees as a condition of approving research; that is not stated in the secondary summary used here.³

AI-enabled medical software is governed separately by the Central Drugs Standard Control Organisation (CDSCO). MediaNama reports that CDSCO already regulates such software under the Medical Devices Rules, 2017, and that its guidance on medical device software, which went through a draft of 21 October 2025 and a final version in 2026, requires developers to disclose how AI models work and what datasets were used to train and test them, to address bias, explainability, model drift and hallucinations, to monitor systems after launch, and to justify whether models trained outside India suit Indian clinical settings.⁷⁶ The ICMR guidelines and the CDSCO guidance therefore address different questions. ICMR's concerns the ethics of research and care, and CDSCO's the licensing of products.

The India AI Governance Guidelines of November 2025 list ICMR's May 2023 AI guidelines among the earlier AI policy documents, and recommend that sectoral regulators enforce approaches suited to their domains.⁴ MediaNama later noted that MeitY's AI Governance and Economic Group (AIGEG), constituted in April 2026, does not include ICMR even though MeitY's own guidelines had recommended including it.⁶⁵

On 6 August 2026 MediaNama reported that the Government told the Lok Sabha on 4 August 2026 that CDSCO has not adopted AI for regulatory approvals.⁷ On 18 September 2026 the Drugs Controller General of India said that CDSCO had formed an expert committee on AI, discussing how to evaluate, approve and regulate AI in drug discovery. MediaNama found no published order, membership or terms of reference.⁶ That committee's remit therefore appears to be AI in drug development, not the clinical and research ethics ground covered by ICMR's guidelines.

A continuing gap is that there is no statute or binding rule that makes the ICMR principles mandatory for developers outside regulated research settings.

As of 15 October 2026 the ICMR guidelines remain published on ICMR's website and no revised edition was identified in the sources reviewed.¹² They remain non-binding. Binding obligations on health AI arise instead from sector laws, including the Medical Devices Rules, 2017, the DPDP Act, 2023 (whose main obligations commence on 13 May 2027), and professional-ethics rules.

The full ICMR PDF was too large to be retrieved for this entry, so the descriptions of its contents come from ICMR's web page and MediaNama's summary and should be checked against the original text.