SAHI and BODH: what India's health AI strategy asks of hospitals

On 17 February 2026, at the India AI Impact Summit, the Ministry of Health and Family Welfare launched SAHI, a national strategy for AI in healthcare, and BODH, a platform for benchmarking health AI models on Indian data before deployment. Together they set out how AI tools will be classified, tested and held accountable. Hospitals and labs that buy or build AI tools now have a reference to hold vendors against.

What was launched

The Union Health Minister launched SAHI, the Strategy for Artificial Intelligence in Healthcare for India, and BODH, the Benchmarking Open Data Platform for Health AI, on 17 February 2026 during the summit, which ran from 16 to 20 February in New Delhi. PIB describes SAHI as a governance framework, policy compass and national roadmap for responsible AI in healthcare, and BODH as a structured mechanism for testing and validating AI solutions before deployment at scale. BODH was developed by IIT Kanpur with the National Health Authority.

What SAHI says

Medianama's reading of the strategy document counts 32 recommendations across five pillars. The first pillar, governance, regulation and trust, asks that AI solutions be classified by risk level and regulated accordingly, that training data reflect the population where a tool will be used, that liability be allocated between developer, deployer and user according to who caused the harm, and that tools be monitored after launch for bias and performance problems. The document treats diagnosis, triage, clinical decision-making and treatment as high-risk and routine tasks such as record-keeping as low-risk, though it does not draw a precise line.

SAHI does not create a new law. The ministry points to the DPDP Act as the legal basis for using personal data, to the NHA's Health Data Management Policy and ICMR's ethical guidelines for AI in health for sector rules, and notes that the IndiaAI Governance Guidelines of 2025 found existing laws adequate for most AI risks (Medianama). On the device side, CDSCO issued draft guidelines for software as a medical device, including AI-based software, in October 2025.

Source for the three pillars above: Medianama, 19 February 2026. The strategy has two further pillars not summarised here.

What BODH does

IIT Kanpur describes BODH as an attempt to resolve what it calls the AI quality testing trilemma: reliability, openness and coverage. Developers can benchmark models on real-world Indian datasets without taking raw patient data away, and regulators get a third-party evaluation they did not have to build. Analytics Insight reported that the platform will evaluate systems for accuracy, bias, reliability and adaptability on diverse, anonymised real-world datasets.

The evidence base behind this is real. The ministry's own examples include a cough-based TB screening tool used on over 160,000 people that raised case detection by 12 to 16 percent, and a diabetic retinopathy screening tool launched by AIIMS in December 2025 that lets non-specialist health workers screen from retinal images.

What this means for hospitals and labs

• Ask for the risk class. If a vendor's tool touches diagnosis, triage or treatment, SAHI treats it as high-risk; the vendor should be able to say how it was validated on Indian data and whether it has been through BODH.
• Ask about training data. A radiology or pathology model trained mostly on non-Indian populations is exactly the case SAHI's representative-data recommendation is written for.
• Keep a human sign-off. SAHI's liability recommendation places responsibility on developer, deployer and user according to who caused harm, and the hospital is the deployer.
• Log what the AI did. Post-launch monitoring is a SAHI recommendation, and the same audit trail is what the DPDP Rules require.
• Start with low-risk uses. Record-keeping, coding, scheduling and documentation are where the strategy expects early adoption and where the compliance burden is lightest.

This is the approach we take in Healthixio: AI drafts the discharge summary from a doctor's voice note and the doctor approves it; the model never signs. When a client asks us to build an AI feature that touches clinical decisions, we start with the risk class.

Frequently asked questions

Is SAHI binding on private hospitals?

SAHI is a strategy document with recommendations, not a statute. It relies on existing law, chiefly the DPDP Act, and on sector policies such as the NHA's Health Data Management Policy and ICMR's ethical guidelines for AI. Regulation of AI-based software as a medical device sits with CDSCO, which published draft guidelines in October 2025.

Can a hospital submit a tool to BODH?

BODH is built for AI model providers and evaluators: developers benchmark models on Indian datasets and regulators get independent results. The sources reviewed here do not describe a hospital-facing submission process. The practical route is to ask your vendor whether their model has been benchmarked on Indian data and to request the results in writing.

What counts as high-risk AI under SAHI?

The strategy broadly places diagnosis, patient triage, clinical decision-making and treatment in the high-risk group, and routine tasks such as record-keeping in the low-risk group. It does not publish a precise threshold, so a hospital should treat anything that changes what a clinician decides for a patient as high-risk until told otherwise.