Are AI-Generated Clinical Notes Legally Valid in India?
AI-generated clinical notes can be a valid part of the medical record in India once the treating doctor has reviewed and signed them. We found no Indian law that bars a doctor from using AI to draft a note, and none that gives an unsigned AI draft any standing of its own. The signature is what turns the AI's text into the doctor's record.
This guide is for hospitals and health-tech teams building or integrating AI documentation into their own system. It explains which laws touch the note, and what the software has to do so that the doctor's sign-off can be relied on later. We are software builders, not lawyers, so take your final design to counsel.
Key takeaways
- Electronic records have legal recognition under section 4 of the Information Technology Act, 2000, and section 3A recognises electronic signatures beyond digital signatures (India Code).
- The doctors' code of conduct in force is the 2002 regulations. NMC's 2023 replacement was put in abeyance on 23 August 2023 (BW Healthcare).
- Those regulations require indoor-patient records to be kept for three years from the start of treatment (NMC).
- India's DPDP Rules were notified in November 2025, and most of their obligations take effect 18 months later (PIB).
- After an AI-labelled X-ray report from AIIMS New Delhi went viral in November 2025, commentators said AI reports have no standing without a radiologist's signature (Medical Dialogues).
Are AI-generated clinical notes legally valid in India?
Yes, once a doctor has reviewed and signed them. No Indian statute, National Medical Commission regulation or court ruling we could find deals with AI-drafted clinical notes specifically. The note is judged by the rules that already apply to any medical record, and those rules care about who is responsible for it, not who typed it.
Three things follow from that:
- The record is the doctor's. An AI scribe is a tool for drafting. The treating doctor is responsible for what the signed note says, in the same way as for a note dictated to a junior.
- An electronic note is a real record. Section 4 of the IT Act gives electronic records legal recognition, and section 3A lets a record be authenticated by an electronic signature or a notified authentication technique (India Code).
- An unsigned draft is not a record. It is working material. If the system lets a draft flow into the patient file without approval, the hospital has a problem regardless of how accurate the AI is.
The same view appears outside India. A US healthcare attorney has said an AI draft does not become part of the legal medical record until a clinician reviews, confirms and signs it (Healthcare IT News). Australia's MDA National tells doctors that notes generated by AI are treated as signed off by them (MDA National).
Which Indian laws apply to an AI clinical note?
No single law covers it. Five sets of rules each govern one part of it, and a build has to satisfy all five at once.
| Law or guideline | What it governs | What it means for the note |
|---|---|---|
| IT Act, 2000 (s.3A, s.4) | Electronic records and signatures | An electronic note signed electronically is legally recognised |
| IMC Regulations, 2002 (reg. 1.3) | Doctors' duty to keep and share records | Keep records, in a standard format, and supply them on request |
| Bharatiya Sakshya Adhiniyam, 2023 (s.63) | Electronic evidence in court | A printout of the note needs a certificate, including a hash value |
| DPDP Act, 2023 and Rules, 2025 | Personal data, including health data | Consent for recording and for each further use; retention and deletion |
| Consumer Protection Act, 2019 | Patients' claims for deficient service | The signed note is the evidence in a negligence complaint |
The ICMR's 2023 ethical guidelines for AI in healthcare sit alongside these. They are guidance, not law, and they include a principle on accountability and liability, with regular audits of AI systems (ICMR).
Does the doctor have to sign every AI-generated note?
In practice, yes. The regulations make the doctor responsible for the record, and an unsigned AI note has no author anyone can hold to it. The AIIMS episode shows how it reads in public: a chest X-ray report labelled "Verified/AI-Radiologist" drew criticism that AI reports have no standing without a radiologist's signature, and AIIMS said its AI findings were preliminary and for clinicians' use only (Medical Dialogues). Those were commentators' views, not a ruling, but no one argued the other side.
For the software, "signing" has to mean something. A button labelled "Approve" that a nurse can press on the doctor's behalf is not a signature. The doctor has to be logged in as themselves, and the system has to record that this person approved this version at this time.
Is patient consent needed to record a consultation?
Yes, and get it separately from consent to treatment. Under the DPDP Act, consent is the default basis for processing health data and must be free, specific, informed and unambiguous (K&S Partners). A signature on the admission form is not blanket authority for later uses such as AI training (KPMG).
The timing matters for anyone starting a build now. The Rules were notified in November 2025, and their main obligations come into force 18 months later, in May 2027 (PIB; Legal500). A system started today will go live under them, so design for them now rather than retrofitting.
What the consent step needs, at minimum:
- A plain-language notice that the consultation will be recorded and drafted by AI.
- A separate choice for any use beyond the note itself, such as training or quality review.
- A record of who agreed, when, and to which version of the notice.
- A way to refuse, after which the doctor writes the note as before.
What counts as the medical record: the note, the transcript or the audio?
The signed note is the record. The audio, the transcript and the AI's first draft are working material, and each needs its own retention rule. This is the part most guides skip, and it is where builds go wrong.
The 2002 regulations require indoor-patient records to be kept for three years from the start of treatment, and to be supplied when the patient or a legal authority asks (NMC). That applies to the signed note, and hospitals often keep records much longer for their own reasons. It does not oblige you to keep every recording for years.
| Item | Is it the record? | Sensible default |
|---|---|---|
| Signed note | Yes | Keep for the hospital's full retention period |
| Edit history (AI draft vs final) | Supports the record | Keep with the note, because it shows what the doctor changed |
| Transcript | Working material | Keep for a short, stated period, then delete |
| Audio | Working material | Delete after sign-off unless the patient agreed to longer |
Keeping audio indefinitely feels safe and is not. Under the DPDP Act, data you no longer need is data you should not hold, and a store of recorded consultations is a larger breach target than a store of notes. Whatever you choose, write it down, show it to the patient and make the system enforce it.
Will an AI-drafted note hold up as evidence?
It can, if the system can prove what was signed and that it has not changed since. Electronic evidence in Indian courts now goes through section 63 of the Bharatiya Sakshya Adhiniyam, which replaced section 65B of the Evidence Act from 1 July 2024. The certificate that goes with an electronic record has to state the record's hash value (LiveLaw).
Medical services fall under the Consumer Protection Act, 2019, and negligence complaints usually turn on what the record says. A note that the doctor approved without reading is evidence against them. A note with a clean edit history, showing the doctor corrected the AI's dosage line, is evidence for them.
So the build has to support the evidence trail:
- Version every note. Keep the AI draft and each edit, not just the final text.
- Lock after signing. A signed note can be amended only by an addendum that is itself signed and timed.
- Log access. Record who opened, changed, printed or exported the note.
- Export with integrity. Produce a copy with its hash so the hospital can complete a section 63 certificate without a developer.
What should the software never let AI do?
Never let the AI sign, and never let an AI draft reach the patient, a pharmacy or another clinician before a doctor approves it. Those two rules cover most of the risk. The others follow from them:
- No prescriptions issued from an unapproved draft.
- No silent overwrite of a signed note by a later AI run.
- No "approve all" button for a batch of notes.
- No training on recordings or notes without the separate consent above.
Building it or integrating it: what we look at first
This is a build and integration question, not a product question. If you are choosing an AI scribe product, the vendor should answer each point above for its own system. If you are building AI documentation into your own hospital system, or connecting one to it, these become your requirements.
We look at four things first:
- Identity. Does every doctor have a personal login strong enough to stand as an electronic signature?
- Note states. Does the records module have draft, signed and amended states, or only "saved"?
- Audit log. Is there a log of who did what, and can the hospital read it without the vendor?
- Retention. Can the system delete audio and transcripts on a schedule while keeping the signed note?
If any answer is no, fix it before adding AI, because the AI makes each gap larger. Our guide to what it costs to build AI clinical documentation into a hospital system prices this work, and the AI feature development cost guide covers the model side.
What we've seen at DECK IT
DECK IT has built software from Ahmedabad since 2017, including our own hospital management software. Our sister company Slaya Technologies owns MedSlay, an AI clinical documentation product for Indian clinics, so we have watched these questions come up in a real build rather than in theory.
The lesson was that the sign-off screen deserved more design time than the AI. Doctors approve what is in front of them quickly, so the screen has to show what the AI changed or inferred, and make the doctor look at the parts that matter, such as medicines, doses and allergies. A legally sound process that doctors click through without reading protects nobody.
The second lesson was about records modules that only have a "save" state. Adding draft and signed states to an old system is often the largest piece of work, and it has nothing to do with AI. Our hospital software features guide covers what the clinical records module should already do.
Frequently asked questions
Are AI-generated clinical notes legally valid in India?
Yes, once the treating doctor has reviewed and signed them. We found no Indian law that bars AI drafting or that gives an unsigned AI note standing of its own. The signed note is judged like any other medical record.
Does a doctor have to sign an AI-generated note?
Yes. The doctor is responsible for the medical record, and an unsigned AI draft has no author who can be held to it. The system should record which doctor approved which version and when.
Is patient consent needed to record a consultation for an AI scribe?
Yes. Under the DPDP Act, consent is the default basis for processing health data and must be specific and informed. Take it separately from treatment consent, and take a further consent for any use such as AI training.
Should the audio recording be kept as part of the medical record?
No, the signed note is the record. Audio and transcripts are working material, and a sensible default is to delete audio after sign-off unless the patient agreed to longer. Write the rule down and have the system enforce it.
How long must medical records be kept in India?
The 2002 regulations require doctors to keep indoor-patient records for three years from the start of treatment. Many hospitals keep them longer, so check your own policy and any rules for your state or accreditation.
Can an AI-drafted note be used as evidence in court?
Yes, like any electronic record. It needs a certificate under section 63 of the Bharatiya Sakshya Adhiniyam that states the record's hash value, so the system should be able to export a note with its hash and edit history.
Who is responsible if an AI note is wrong?
The doctor who signed it, in the first instance, because the signed note is their record. A hospital can also face claims over the systems it chose and how staff were trained, so build review screens and audit logs that show reasonable care.
Conclusion
AI-generated clinical notes are valid in India on one condition: a doctor reads and signs them. Build the system around that condition with draft and signed states, a real personal sign-off, an edit history, separate consent and a retention rule for audio. Then put the design in front of your legal counsel before go-live.
If you are adding AI documentation to a hospital system and want the records side checked first, tell us which system you run.