AI Healthcare Regulations in India: Laws and Compliance
AI Healthcare Regulations in India: Laws and Compliance
India has no single AI healthcare law, but rules still apply. CDSCO’s 2026 medical device software guidance, the Medical Devices Rules, the DPDP Rules and ICMR ethics guidelines together shape compliance. This article explains what regulators expect from AI diagnostic tools, how approvals really work, and how to protect patient data.
Introduction
India’s healthtech sector has a habit it can no longer afford: building first and asking about regulation later.
For years, the order of operations has been the same. Train the model, polish the demo, chase the pilot, and deal with compliance once someone asks. Someone now asks, and early. Hospital buyers increasingly ask for a CDSCO licence before they will talk about a pilot, and an impressive accuracy score does not substitute for one.
That is the real shift behind AI healthcare regulations in India. With CDSCO’s final medical device software guidance in place, the rules are clearer, the expectations are higher, and the excuse of ambiguity is gone. Compliance is no longer paperwork at the end of the roadmap. It is the entry ticket.
Quick answer: is AI in healthcare regulated in India?
Yes, but through a patchwork rather than one dedicated AI law. Software with a medical purpose is regulated as a medical device by CDSCO under the Medical Devices Rules, 2017. Patient data falls under the Digital Personal Data Protection (DPDP) framework. Ethical use is guided by ICMR’s 2023 guidelines. You need to satisfy all three.
Which laws make up AI healthcare regulations in India?
Think of four layers:
- Medical Devices Rules, 2017 (MDR): decide whether your AI needs a licence and of what kind.
- CDSCO’s Guidance Document on Medical Device Software: explains how MDR applies to software and AI.
- DPDP Act, 2023 and DPDP Rules, 2025: govern how you collect, use and protect patient data.
- ICMR Ethical Guidelines (2023): set expectations for fairness, consent and accountability.
There is no standalone “AI Act” for healthcare yet. Regulators are applying existing laws to AI, which is why the details matter so much.
CDSCO AI medical device regulations: does your software count?
The key document is recent. On July 21, 2026, CDSCO issued its final Guidance Document on Medical Device Software (MDSW) under the Medical Devices Rules, 2017, following an earlier draft released for stakeholder consultation in October 2025.
The guidance covers AI medical software regulations in India directly. It applies to software that performs medical functions on its own or works alongside connected devices, including AI platforms that monitor chronic disease and support clinical decisions. Examples include a tool that screens chest X-rays for tuberculosis, an app that interprets ECG readings, and cloud-based clinical decision support.
What is excluded? Software used purely for general wellness, fitness tracking or ERP functions falls outside the definition. The test is intended purpose. A step counter is not a device. A tool that suggests a diagnosis probably is.
Classification also follows risk. Software that drives or influences a hardware device should sit in the same risk class as that hardware. Higher risk means more evidence and closer scrutiny.
One honest caveat: the guidance is not itself binding. It interprets the binding rules. In practice, though, reviewers use it as their yardstick, so treat it as mandatory.
CDSCO approval requirements for AI medical devices
The regulatory requirements for AI diagnostic software in India depend on your risk class and whether you manufacture or import. Across all of them, reviewers want to see the same building blocks:
- A clear intended-use statement. Who uses it, on which patients, for what decision. Vague claims cause the most delays.
- Risk classification. Justify the class with reference to CDSCO’s framework.
- A quality management system. Software development must follow documented, controlled processes.
- Technical documentation. This covers architecture, validation, cybersecurity and labelling.
- Clinical or performance evidence. Proof that the AI works for its intended use.
- A post-market plan. How you will watch the product once it is live.
Dataset evidence deserves extra attention. The guidance emphasises that AI software must perform appropriately across the populations, care settings and environments in its intended use. A model trained only on data from one metro hospital will be questioned. Existing AI products should be reviewed for dataset quality, bias, generalisability and relevance to Indian patients.
You must also disclose your data sources. The device master file needs to state whether training data was real-world data from public datasets or literature, or artificially generated.
Medical Device Rules requirements for AI software in India: what about model updates?
This is the question every AI team asks, because models improve through retraining.
Industry pushed for relief. Bodies like the Medical Technology Association of India (MTaI) had questioned how frequent updates and learning loops would work if every change needed a fresh licence. The answer is an Algorithm Change Protocol (ACP), a pre-defined plan describing how and when your model may change.
Do not read that as a free pass, though. Even with an ACP, model updates still need approval or notification to the regulator. The practical lesson: a validation package frozen at one software version will not satisfy a product that retrains or updates. Build version control and change documentation from day one.
What happens after launch?
Approval is not the finish line. Manufacturers must monitor for patient harm, reduced effectiveness, non-conformities and cybersecurity vulnerabilities, and AI software must specifically be tracked for performance, bias and drift. If something goes wrong, corrective action can include recalls.
In plain terms: if your model quietly gets worse on a new patient population, you are expected to notice before a clinician does.
Health data protection India: how does the DPDP Act affect healthcare AI?
Healthcare AI runs on sensitive data, so this is where many teams carry the biggest risk.
The DPDP Rules, 2025 were notified on 13 November 2025. Organisations have an 18-month runway to full compliance, falling in May 2027. A few things worth knowing:
- Consent must be informed and verifiable. Hidden tick boxes will not hold up.
- Breach reporting is fast. The Rules require reporting to the Data Protection Board, with a detailed report within 72 hours.
- Larger players face extra duties. Large-scale processing of health data is one factor that can lead the government to designate an organisation a Significant Data Fiduciary, which must then appoint an India-based Data Protection Officer.
- Children’s data gets special handling. Healthcare providers have a limited exemption, only for protecting the child’s health. It is not a general licence to process.
- Consent managers are emerging. They must be Indian companies, which may reshape how health apps collect permissions.
This is the core of patient data protection for healthcare AI in India. Training a model on patient records counts as processing. Anonymisation, purpose limits and retention rules all apply, so “we only used it for research” is not a defence.
Where do ICMR guidelines fit?
ICMR’s 2023 guidelines are not a licence regime, but ethics committees use them to review AI research, and hospitals increasingly use them to vet vendors. They set out ten principles, including accountability, autonomy, data privacy, safety, equity, fairness and validity.
Two points matter most for product teams. Patients should be able to refuse AI involvement in their care, and AI tools need clinical and field validation before use. A “human in the loop” is expected, not optional.
Healthcare AI compliance India: a practical checklist
Use this to find your gaps:
- Write a one-paragraph intended-use statement, then check it against the MDSW definition.
- Confirm your risk class before you build the evidence plan.
- Audit training data for Indian representativeness and document its sources.
- Set up a quality management system and a change-control process.
- Draft your Algorithm Change Protocol early.
- Map every flow of patient data, then match each one to a consent and a retention rule.
- Plan breach response now, not after an incident.
- Prepare a post-market monitoring plan covering bias and drift.
- Keep an ethics review trail for any research use.
Frequently asked questions
Do all healthcare AI tools need CDSCO approval?
No. Only software with a medical purpose, such as diagnosis, monitoring or treatment decisions, is regulated as a device. General wellness apps are excluded.
Is there a specific AI law for healthcare in India?
Not yet. Existing laws, mainly the MDR 2017 and the DPDP framework, are applied to AI.
When do DPDP obligations fully apply?
The main compliance duties take effect in May 2027, following the 18-month transition from the November 2025 notification.
Can I update my AI model without a new approval?
Only within an approved Algorithm Change Protocol, and notification or approval to CDSCO may still be required.
Do ICMR guidelines carry legal force?
They are ethical guidance rather than law, but ethics committees and hospital buyers rely on them heavily.
The takeaway
India’s approach is practical: regulate by purpose, demand evidence, and keep watching after launch. Teams that treat compliance as part of product design, not paperwork at the end, move faster because they are rarely forced to rebuild.
This article is general information, not legal advice. Check current CDSCO and MeitY notifications or consult a regulatory professional before filing.