A recommendatory framework, adopted as binding
SAHI is a national guidance framework rather than a statute. It does not replace India's existing data law; it works alongside the Digital Personal Data Protection Act 2023, the National Health Authority's Health Data Management Policy, and ICMR's ethical guidelines for AI in health research.
Being recommendatory, it can be ignored. The programme's position is that it will not be. A platform asking government agencies and international funders to trust it with disaster response cannot credibly treat the national health AI framework as optional — and the frameworks that follow recommendatory ones are rarely more permissive.
SAHI is also notable for what it does not say. Its third governing principle holds that responsible innovation should be prioritised over cautionary restraint where other things are equal — a markedly different posture from the precaution-first framing dominant elsewhere. The programme reads that as licence to deploy, not licence to skip validation.
Five pillars, and what each requires of RISE
| Pillar | What SAHI asks | How RISE answers |
|---|---|---|
| Governance and evidence generation | Risk-proportionate governance; dedicated AI units or nodal cells; clearly defined roles for humans and AI with oversight and escalation processes; validation before clinical use. | Every AI component risk-classified (CMP-05); an AI oversight function with a named accountable owner; human-in-the-loop enforced in the orchestrator rather than by policy; no AI output reaches a clinical or dispatch decision unreviewed. |
| Data and digital infrastructure | Safe, ethical, robust and transparent digital and data foundations; interoperability with national digital health infrastructure. | ABDM conformance (CMP-01); FHIR R4 to NRCeS profiles; data residency in India; separation of governed clinical storage from ledger-anchored proof. |
| Workforce readiness | A future-ready health workforce; a FRAC-based approach to roles, activities and competencies; digital literacy. | SYS-08 Training and Support treated as core platform functionality; role-based curricula mapped to FRAC; competence records that follow the person between deployments; specialised air ambulance interaction training. |
| Research and evidence | Evidence generation standards; training data reflecting the population and setting where the tool will operate. | BODH benchmarking before deployment (CMP-03); model calibration to local hazard, terrain and population profile; published accuracy from historical backtesting; governed anonymised research access. |
| Ecosystem enablement and market stewardship | Addressing fragmented, opaque or narrowly cost-driven procurement that incentivises stop-gap solutions and vendor lock-in. | Open documented APIs and published deprecation policy; no requirement to replace existing estates; data ownership retained by the deploying organisation; exportable records in standard formats. |
The seven sutras
SAHI codifies seven governing principles, which it calls sutras. The programme's reading of each, as it applies to a disaster and emergency platform, is set out below. Where our interpretation is contestable we have said so rather than smoothing it over.
- Public interest and trust
- AI adoption anchored in public interest and long-term system resilience. For RISE this constrains the revenue model as much as the technology: freemium access and tiered pricing exist so that the organisations closest to affected populations are not priced out of the safety infrastructure.
- AI as assistant, not decision-maker
- AI assists rather than replaces clinicians and health workers. Every predictive output in RISE is advisory. The platform recommends a dispatch; a named authoriser commits it, with the reason recorded.
- Benefit maximisation over cautionary restraint
- Responsible innovation prioritised over precaution where other things are equal. In disaster response the cost of inaction is measurable and often fatal, which makes this principle unusually load-bearing here. It is not, however, a reason to deploy an unvalidated model.
- Equity and inclusion
- Prioritising underserved and rural populations. This is the founding premise of Cad Care Aid Foundation and therefore of the platform behind it — voice and SMS request paths, offline operation, and multilingual interface exist for exactly this reason.
- Transparency and explainability
- AI behaviour that can be inspected and explained. Forecasts are published with confidence intervals; every AI action is written to an immutable log with its inputs, outputs and rationale.
- Accountability
- Clear allocation of responsibility when something goes wrong. Actor attribution covers delegated and system actions; override decisions capture a mandatory reason; audit trails are reconstructible by incident.
- Risk-proportionate regulation
- AI solutions classified by risk level and regulated accordingly. See CMP-05 for the classification applied to each RISE component.
Independent commentary has noted that SAHI's thirty-two recommendations arrive without an identified budget, funding source or implementation allocation, and without clear milestones or measurable targets. That is a real limitation of the framework and it affects planning: a platform cannot align to deadlines that have not been set. We track the framework as it develops rather than treating the February 2026 document as final.