India's health-AI framework puts hospital governance on the agenda

On 17 February 2026, the Ministry of Health and Family Welfare launched SAHI, the Strategy for Artificial Intelligence in Healthcare for India, and BODH, the Benchmarking Open Data Platform for Health AI. The latter was developed by IIT Kanpur with the National Health Authority to support evaluation of AI models using real-world health datasets.

Start with the tools, then assign responsibility

Lifeline's management view: a useful response begins with an inventory. List each AI-enabled tool approved for use, including features inside existing hospital software. Record its task, the information it receives, the person reviewing its output and the supplier responsible for support.

The inventory should distinguish a tool that helps arrange appointments from one that influences a clinical decision. Their consequences differ, so the scrutiny and authority needed to approve them should differ too. Ask the clinical lead to define the boundary wherever patient care may be affected.

For example, suppose a hospital is evaluating an assistant that prepares a draft clinical document. A review should establish which professional checks it, how corrections are recorded and whether the final signed record can be traced back to its source information. An attractive draft is only one part of that workflow. This is an illustrative management exercise, not a report of a deployment or a recommendation for a particular product.

What this means for your hospital

Ask what the evidence actually covers. Request the intended use, evaluated settings and limitations for the exact product under consideration. Ask how the supplier knows it is suitable for your hospital's workflow. A benchmark result is an input to that discussion; it should not be presented to your team as blanket government approval.

Keep the approval small enough to inspect. State the department, permitted task, authorised users and review period in the approval record. Name the person who can pause use. A limited, supervised introduction gives that owner a defined set of outputs to check before deciding whether the scope should grow.

Decide what happens when the tool changes. Ask the vendor how it identifies model or software versions, communicates material changes and preserves your records. Write down which changes require another hospital review. Include an exit route that your staff can use without losing access to the information they need.

Bring the results back to a responsible meeting. Review reported errors, staff corrections and the work required to supervise the system. Give clinicians, operations and technology staff a shared record of the decision to continue, change or stop. The purpose is a decision that somebody owns, rather than an approval that remains open indefinitely.

These are proposed hospital practices, not a restatement of legal duties. They belong alongside a technology and data review. When a purchase is proposed, put the operating case next to the oversight plan so that usefulness and accountability are assessed together.