Which Indian AI rules actually bind you
India has no AI statute. This separates what binds you from the report, the voluntary guidelines and the draft that are cited as though they did.
What binds you
Pick your entity type. The answer names every Indian AI-facing instrument that reaches you, ordered by what it can actually do to you rather than by when it was published — and it says plainly where the answer is that nothing AI-specific binds you at all, which it is for 4 of the 7 entity types here.
A broker, portfolio manager, merchant banker, AIF manager, RTA or any other person regulated by SEBI.
One binding AI rule reaches you, and it is unusually wide: sole responsibility for the output of AI tools, including tools bought from a third party, at any scale of adoption.
| Instrument | Force | From |
|---|---|---|
| CERT-In Directions, 2022 No. 20(3)/2022-CERT-In | Binding | 27 June 2022 |
| SEBI (Intermediaries) (Amendment) Regulations, 2025 — regulation 16C SEBI/LAD-NRO/GN/2025/226 | Binding | 10 February 2025 |
| Digital Personal Data Protection Rules, 2025 G.S.R. 846(E) | Binding | 13 May 2027 |
| MeitY India AI Governance Guidelines | Voluntary | 5 November 2025 |
| RBI FREE-AI committee report | A report to a regulator | 13 August 2025 |
| SEBI consultation paper on guidelines for responsible AI/ML use | Draft | 20 June 2025 |
What each one asks for, and what it can do to you
- CERT-In Directions, 2022 — binding. A cyber incident reported to CERT-In within six hours of noticing it. The Directions are technology-neutral, so an incident caused by or involving an AI system is reportable on exactly the same terms as any other.
Enforceable against you today. A breach is actionable by the regulator under its own powers. Not an AI instrument, and included because it is the one obligation on this page that is certain to apply to an AI incident today. An organisation waiting for AI-specific reporting rules already has reporting rules. - SEBI (Intermediaries) (Amendment) Regulations, 2025 — regulation 16C — binding. A person regulated by SEBI using AI or machine learning tools — built in-house or procured from a third-party technology service provider, and irrespective of the scale and scenario of adoption — is solely responsible for the privacy, security and integrity of investor and stakeholder data, for the output arising from those tools, and for compliance with applicable laws.
Enforceable against you today. A breach is actionable by the regulator under its own powers. The words that do the work are “irrespective of the scale and scenario of adoption”. There is no pilot exemption and no threshold below which the responsibility does not attach, and procuring the tool does not move it to the vendor. - Digital Personal Data Protection Rules, 2025 — binding. Where an AI system processes digital personal data, the Data Fiduciary obligations — notice, security safeguards, breach intimation, retention limits and Data Principal rights — reach that processing like any other. They commence on 13 May 2027.
Enforceable against you today. A breach is actionable by the regulator under its own powers. The date matters more here than elsewhere, because model training on personal data is the use most often described as already regulated in India. It is not yet: rules 3 and 5 to 16 commence eighteen months after publication. - MeitY India AI Governance Guidelines — voluntary. A set of governance recommendations for AI development and deployment in India, addressed to industry and to government.
Not enforceable. The document says so of itself, and adopting it is a choice rather than compliance. The Guidelines say of themselves that voluntary frameworks lack legal enforceability or punitive action, and they recommend amending the Information Technology Act rather than passing an AI law. They are the document most often cited as evidence that India “has AI governance rules”. - RBI FREE-AI committee report — a report to a regulator. A framework for the responsible and ethical enablement of artificial intelligence, recommended to the Reserve Bank by the committee that wrote it.
A committee reporting to a regulator, not the regulator instructing you. It may become policy and has not. A committee reporting to the RBI is not the RBI instructing a regulated entity. FREE-AI is routinely written up as “the RBI’s AI framework”, which reverses the direction of the document. Nothing in it is enforceable against a bank today. - SEBI consultation paper on guidelines for responsible AI/ML use — draft. Proposals for the responsible usage of AI and machine learning in the Indian securities markets, issued for comment.
A consultation paper. Nothing in it binds anyone, and the proposals in it may change or be dropped. This is the source of the “kill switch” requirement that has since been reported as though SEBI had imposed one. It is a consultation paper. SEBI’s final guidelines on responsible AI/ML use were confirmed still unissued on 3 September 2026, against SEBI’s own circular and regulation listings rather than against commentary.
Indicative, and not legal advice. Every binding instrument here was read from the issuing regulator’s own notification; the non-binding documents are linked to their publishers. Whether an instrument reaches your organisation is a determination for your legal team.
Take this away as a print-ready AI obligations brief
Everything above is yours already. This is the version you can put in front of your team — the same result with your details on it, the owner columns your incident file needs, and the checklist attached. It opens here and lands in your inbox, so it is to hand when you actually need it rather than in a tab you closed.
Five documents, five different forces
India has no AI statute and no AI regulator. What it has is a set of documents of very different weight, published within eighteen months of each other, which are routinely summarised together as “India’s AI rules”. Sorting them is most of the work.
| Force | Instruments | What it means |
|---|---|---|
| Binding | 5 | Enforceable against you today. A breach is actionable by the regulator under its own powers. |
| Advisory, largely drafted as “shall” | 1 | Issued as a circular, in force, and expressly diluting no other obligation — but titled an advisory, and encouraging compliance on its covering page. Read the annexure rather than the label: what is drafted with “shall” is what a supervisor will ask about. |
| Voluntary | 1 | Not enforceable. The document says so of itself, and adopting it is a choice rather than compliance. |
| A report to a regulator | 1 | A committee reporting to a regulator, not the regulator instructing you. It may become policy and has not. |
| Draft | 1 | A consultation paper. Nothing in it binds anyone, and the proposals in it may change or be dropped. |
The date that is reported wrongly
All three SEBI amendments are dated 6 February 2025 and were published in the Gazette on 10 February 2025, and each says it comes into force on the date of publication. The depositories amendment, SEBI/LAD-NRO/GN/2025/225, is nonetheless widely reported as commencing on 1 April 2025.
Its own regulation 2 is why. That regulation sets 1 April 2025 for the amendment generally, and then carries a proviso putting “the amendment in regulation 3(IV)” into force on gazette publication instead. Regulation 3(IV) is the AI clause. So the AI provision commenced on 10 February 2025 with the other two, and the 1 April date is true of the fee clauses the same amendment carries and false of the clause everybody is citing it for — a difference of seven weeks on the only binding AI obligation most Indian firms have.
The kill switch is still a draft
The requirement to build a kill switch into an AI system comes from SEBI’s consultation paper on guidelines for the responsible usage of AI and ML, issued for comment on 20 June 2025. It has since been reported in several places as though SEBI had imposed one.
SEBI’s final guidelines were confirmed still unissued on 3 September 2026, checked against SEBI’s own circular and regulation listings rather than against commentary. Building to a draft is a reasonable engineering decision and it is not compliance with anything, and the two are worth keeping apart in a board paper.
- Agentic AI in India: who answers when the agent acts — the same instruments read as an argument rather than as a list, including what “the output of a tool” means once the tool takes actions.
- The IFSCA frontier AI advisory — the reference record, with what each of its eleven annexure items actually says.
- Which regulation applies — the technology-neutral instruments that bind whatever you deploy.
Every binding instrument here was read from the issuing regulator’s own notification; the three non-binding documents link to their publishers. Indicative, and not legal advice: whether an instrument reaches your organisation is a determination for your legal team.
Every instrument cited here was verified against the issuing regulator's own notification on .Worked examples
The same calculator, run for a specific entity — with the thresholds and clocks that apply to it, and the instrument each one comes from.
Questions this page answers
- Is there a law regulating artificial intelligence in India?
- No. India has no AI statute and no AI regulator. The binding obligations come from sector regulators using existing powers: SEBI amended three sets of regulations on 6 February 2025, and IFSCA issued a frontier AI circular for GIFT City entities on 4 June 2026. MeitY’s India AI Governance Guidelines of 5 November 2025 are expressly voluntary and recommend amending the Information Technology Act rather than passing an AI law.
- Does the RBI have an AI framework banks must follow?
- No. The FREE-AI document of 13 August 2025 is a committee’s report to the Reserve Bank, not the Reserve Bank instructing a regulated entity, and nothing in it is enforceable against a bank today. It is routinely written up as “the RBI’s AI framework”, which reverses the direction of the document. What does bind an RBI regulated entity is the cybersecurity Directions of 31 July 2026, which apply to the systems it runs whether or not a model is inside them.
- When did SEBI’s AI rule for depositories commence?
- 10 February 2025, on publication in the Official Gazette. The Depositories and Participants amendment is widely reported as commencing on 1 April 2025, but its own regulation 2 carries a proviso putting the amendment in regulation 3(IV) into force on gazette publication instead — and regulation 3(IV) is the AI clause. The 1 April date is true of the fee clauses the same amendment carries and false of the clause it is cited for.
- Has SEBI required a kill switch for AI systems?
- No. The kill switch appears in SEBI’s consultation paper on guidelines for the responsible usage of AI and machine learning, issued for comment on 20 June 2025. It is a draft proposal. SEBI’s final guidelines were confirmed still unissued on 3 September 2026, checked against SEBI’s own circular and regulation listings rather than against commentary.
- What does the IFSCA frontier AI advisory require?
- Despite its title, most of its annexure is drafted with “shall”: frontier AI carried as a named scenario in the cyber risk assessment and put before the Board, a Software Bill of Materials covering open-source components, an inventory of APIs and the applications consuming them with rate-limiting and a whitelist, critical service providers required to furnish evidence of frontier-AI preparedness, monitoring tuned for attack sequences faster than a human could run, and human oversight plus security testing of AI-generated code before production.