Summary
- Between November 2023 and March 2024 MeitY sent intermediaries a series of advisories on deepfakes and AI tools. They did not create a new law. They explained how MeitY expects the existing due diligence duties in the IT Rules, 2021 to be applied to AI.
- The 1 March 2024 AI advisory went furthest: it asked platforms to get the Government's explicit permission before offering "under-tested" or "unreliable" AI models to Indian users and to file an action taken report. Industry pushed back hard.
- The 15 March 2024 revision dropped the permission requirement and the report. In its place: label unreliable models, warn users (for example with a consent pop-up), keep AI from enabling unlawful content, bias or threats to elections, and tag synthetic content with a label or permanent metadata.
- The December 2023 advisory (all intermediaries) and November 2023 advisory (SSMIs) focused on telling users clearly what content is banned and removing deepfakes fast.
- In February 2026 the binding IT Rules Amendment 2026 turned the labelling and metadata idea into law for audio, image and video. The other asks (unreliability labels, bias, elections) remain advisory only.
Who it applies to
- 15 Mar 2024 advisory: "intermediaries and platforms" using or offering AI models, LLMs, generative AI, software or algorithms. Law firms report it was sent to eight large platforms but is framed for all intermediaries (secondary).
- 26 Dec 2023 advisory: all intermediaries (PIB).
- 7 Nov 2023 advisory: significant social media intermediaries (PIB).
- 9 Feb 2026 advisory: intermediaries including AI platforms, AI tool providers and social media intermediaries.
- Foreign companies: yes in practice. The advisories target platforms serving Indian users, such as global chatbot and social media providers.
Key dates
- Advisory to significant social media intermediaries on misinformation and deepfakes (36 hour removal on report)
- Advisory to all intermediaries: communicate Rule 3(1)(b) prohibited content to users at sign-up, login and upload
- AI advisory: explicit government permission before deploying under-tested or unreliable AI; action taken report sought
- Revised AI advisory supersedes 1 March version: permission requirement and report dropped, labelling and metadata kept
- MeitY tells Lok Sabha the AI advisories were issued under the IT Rules, 2021
- Further advisory to AI platforms on religious and other unlawful content
- IT Rules Amendment 2026 in force, codifying labelling and metadata for synthetic audio-visual content
- Draft Second Amendment proposes Rule 3(4) making MeitY advisories part of binding due diligence (not notified as of 2026-10-19)
Full date record
| Date | Event |
|---|---|
| 2023-11-07 | Advisory to SSMIs: identify misinformation and deepfakes, remove reported content within 36 hours (PIB) |
| 2023-12-26 | Advisory to all intermediaries on communicating Rule 3(1)(b) prohibited content (PIB) |
| 2024-03-01 | AI advisory with government permission requirement (official text not located; secondary sources) |
| 2024-03-15 | Revised AI advisory supersedes 1 March version (confirmed in PIB 25 Mar 2026; content per IndiaAI portal and secondary) |
| 2024-07-31 | Lok Sabha reply: AI advisories issued under the IT Rules, 2021 (PIB) |
| 2026-02-09 | MeitY advisory to AI platforms on religious and other unlawful content |
| 2026-02-20 | IT Rules Amendment 2026 (Synthetic Content) in force |
| 2026-03-30 | Draft Second Amendment Rules propose Rule 3(4) (advisories become due diligence); status unconfirmed |
What changed between 1 March and 15 March 2024
| Topic | 1 Mar 2024 | 15 Mar 2024 |
|---|---|---|
| Government permission before deploying under-tested or unreliable AI | Required ("explicit permission") | Dropped |
| Action taken cum status report to MeitY | Required (secondary sources say within 15 days) | Dropped; comply "with immediate effect" (secondary) |
| Label unreliable outputs and use consent pop-ups | Yes | Yes, now the main safeguard for unreliable models |
| No bias, discrimination or threat to electoral integrity | Yes | Yes |
| Label or permanent metadata on synthetic content that could be misinformation or deepfake | Yes | Yes |
| Scope | "Platforms"; a minister's posts said only significant platforms, not startups (secondary, no legal effect) | Framed for all intermediaries and platforms using AI (secondary) |
Penalties and enforcement
- The advisories set no penalty of their own.
- They point to Rule 7 of the IT Rules: an intermediary that fails due diligence loses the Section 79(1) safe harbour and can be prosecuted under the IT Act and criminal law (PIB 7 Nov 2023; PIB 6 Aug 2026).
- Users who create unlawful content face IT Act offences (for example Sections 66C, 66D, 66E, 67, 67A) and BNS offences (PIB 6 Aug 2026).
- If the draft Rule 3(4) is notified, non-compliance with an advisory would itself become a due diligence failure.
What it means for an Indian company
- If you offer a chatbot or GenAI feature to the public, follow the 15 March asks as good practice: a visible "AI can make mistakes" notice, a consent or acknowledgement step, output filters for unlawful content, bias testing, and election-period safeguards.
- For synthetic audio, image and video, do not stop at the advisory. The binding 2026 rules now set the standard (visible label, permanent metadata, no stripping). See IT Rules Amendment 2026 (Synthetic Content).
- For synthetic text, the advisory's labelling ask still applies as guidance, because the 2026 rules cover only audio-visual content.
- No government approval is needed to launch an AI model in India. Anyone quoting the 1 March permission rule is out of date.
- Keep a file of every MeitY advisory you receive and how you responded; it is your best defence if due diligence is questioned.
How the 2026 Amendment Rules Relate
| Advisory ask (2024) | Status after 20 Feb 2026 |
|---|---|
| Label or embed metadata in synthetic content | Codified for audio, image and video in Rule 3(3)(a)(ii) and 3(3)(b); identifier must point to the intermediary's resource. Text stays advisory only. |
| Identify the user or first originator via metadata | Not carried over as such; the rules require an identifier of the computer resource, and identity disclosure to victims under Rule 3(1)(ca)(ii)(III). |
| Do not enable unlawful content via AI | Codified and expanded in Rule 3(3)(a)(i) with named categories. |
| Tell users about consequences | Codified and tightened: quarterly notices (Rule 3(1)(c)) plus SGI warnings (Rule 3(1)(ca)). |
| Remove deepfakes within 36 hours | Superseded by 3 hour, 2 hour and 36 hour clocks in Rules 3(1)(d) and 3(2). |
| Label unreliable models, consent pop-ups | Not codified; remains advisory. |
| No bias, no threat to electoral integrity | Not codified as a specific duty; remains advisory. |
| Government permission for untested AI | Withdrawn on 15 Mar 2024; never codified. |
Obligations
Showing 11 of 11
| ID | Obligation | Reference | Applies to | Controls |
|---|---|---|---|---|
| MEITY-ADV-01 | Make sure AI models, LLMs, GenAI tools, software or algorithms on your platform do not let users host or share unlawful content under Rule 3(1)(b) or break the IT Act or other laws. | 15 Mar 2024 advisory (para not verified); IndiaAI portal; links to Rule 3(1)(b) | Both | |
| MEITY-ADV-02 | Make sure your systems, including AI, do not allow bias or discrimination. | 15 Mar 2024 advisory (para not verified); IndiaAI portal | Both | |
| MEITY-ADV-03 | Make sure your systems, including AI, do not threaten the integrity of the electoral process. | 15 Mar 2024 advisory (para not verified); IndiaAI portal | Both | |
| MEITY-ADV-04 | Offer under-tested or unreliable AI models to Indian users only after labelling that their output may be unreliable or fallible. | 15 Mar 2024 advisory (para not verified); IndiaAI portal | Both | |
| MEITY-ADV-05 | Use a consent pop-up or similar step that tells users plainly about the possible unreliability of AI output. | 15 Mar 2024 advisory (para not verified); IndiaAI portal | Both | |
| MEITY-ADV-06 | Tell users through terms of service and user agreements what happens if they deal in unlawful content: removal, account suspension or termination, and punishment under law. | 15 Mar 2024 advisory (para not verified, secondary); restates Rule 3(1)(c) | Both | |
| MEITY-ADV-07 | Where your tool creates or modifies text, audio, image or video that could be misinformation or a deepfake, label it or embed a permanent unique metadata or identifier that shows it came from your resource and can identify the user or first originator. | 15 Mar 2024 advisory (para not verified); IndiaAI portal and secondary | Both | |
| MEITY-ADV-08 | State Rule 3(1)(b) prohibited content clearly in terms of service, and remind users at first registration, at every login and when uploading or sharing. | 26 Dec 2023 advisory (as quoted by PIB) | Both | |
| MEITY-ADV-09 | Tell users which penal provisions (IT Act, criminal code) apply to violations, and that the platform must report legal violations to law enforcement. | 26 Dec 2023 advisory (as quoted by PIB) | Both | |
| MEITY-ADV-10 | SSMIs: make reasonable efforts to identify misinformation and deepfakes, and remove reported deepfakes within 36 hours. | 7 Nov 2023 advisory (as summarised by PIB) | Deployer | |
| MEITY-ADV-11 | AI platforms: build safeguards (algorithmic filters, moderation, human oversight, culturally informed review) so AI does not generate distorted religious or other unlawful content, and review training data, outputs and grievance processes. | 9 Feb 2026 advisory, paras 4 and 5 | Both |
Controls that do the most work here
- SC-AI-034 Unlawful and prohibited content safeguards4 obligations
- SC-AI-025 Public AI transparency statement and notices3 obligations
- SC-AI-036 Deepfake and impersonation safeguards3 obligations
- SC-AI-022 Disclose AI interaction to users2 obligations
- SC-AI-010 Training and input data governance (provenance, quality)1 obligations
- SC-AI-013 Data bias and representativeness checks1 obligations
Evidence an auditor will ask for
- A register of MeitY advisories received, with owner, gap assessment and closure date.
- Screenshots of reliability disclaimers and consent or acknowledgement screens for each AI feature.
- Pre-release test reports for "under-tested" judgments: what testing was done and why a disclaimer was or was not added.
- Bias and fairness test results, and any election-period content safeguards (for example restrictions on candidate deepfakes).
- Output filter and safety policy documentation mapped to Rule 3(1)(b) categories.
- Terms of service versions and logs showing reminders at sign-up, login and upload.
- Labelling and metadata specifications for synthetic text and media, and tests showing they persist.
- Takedown logs for deepfake reports (36 hour target in the 2023 advisory; now superseded by the 2026 rule timelines).
Common mistakes
- Believing India requires government permission to launch an AI model. That was the 1 March 2024 text and it was withdrawn two weeks later.
- Treating the advisories as dead after the 2026 rules. MeitY still cited the 26.12.2023 and 15.03.2024 advisories in a March 2026 Rajya Sabha reply.
- Treating the advisories as binding law with their own penalties. Their force comes from the underlying IT Rules.
- Labelling only images and video. The 2024 advisory also mentions synthetic text that could be misinformation.
- Quoting "issued to eight platforms only" as an exemption. The text is framed for all intermediaries using AI (secondary sources).
- Assuming the old 36 hour deepfake removal target is still the limit. The 2026 rules cut some clocks to 3 and 2 hours.
Glossary
- Advisory: a formal letter from a ministry explaining how it expects existing rules to be followed; not itself a statute or rule.
- Intermediary: a service that hosts or carries content for others (social network, marketplace, AI chat platform).
- Under-tested or unreliable AI: the advisory's undefined phrase for models whose outputs may be wrong; law firms note it lacks a test standard.
- Consent pop-up: an on-screen message the user must acknowledge, here used to warn that AI output may be unreliable.
- Deepfake: synthetic media that makes a real person appear to say or do something they did not.
- Permanent unique metadata or identifier: a tag embedded in content that cannot be removed and points to its origin.
- Safe harbour: the Section 79 IT Act shield against liability for user content, lost if due diligence fails.
- SSMI: significant social media intermediary, above 50 lakh registered users in India (PIB).
Key takeaways
- The March 2024 advisories are guidance issued under the IT Rules framework, not standalone law; their teeth come from the safe harbour rules.
- 1 March asked for government permission to deploy unreliable AI; 15 March dropped that and kept labelling, user warnings and anti-bias asks.
- The 2026 amendment codified the labelling and metadata part for audio-visual content and replaced the old takedown timelines.
- Unreliability disclaimers, consent pop-ups, bias and election safeguards remain advisory but are still cited by MeitY in 2026.
- The official text of the March 2024 advisories was not found on a government site; treat paragraph-level detail as unverified until a copy is obtained.
Official sources
- PIB, "MeitY issues advisory to all intermediaries to comply with existing IT rules", 26 Dec 2023, accessed 2026-10-19
- PIB, advisory to SSMIs on misinformation and deepfakes, 7 Nov 2023, accessed 2026-10-19
- PIB, Lok Sabha reply on deepfakes and AI advisories, 31 Jul 2024, accessed 2026-10-19
- PIB, Rajya Sabha reply citing advisories of 26.12.2023, 15.03.2024 and 29.12.2025, 25 Mar 2026, accessed 2026-10-19
- PIB, "Government Strengthens Regulatory Framework to Address AI-Generated Deepfakes", 6 Aug 2026, accessed 2026-10-19
- IndiaAI portal, "Revised AI Advisory: How AI Regulation Can Impact India" (March 2024), accessed 2026-10-19 (government portal article, not the advisory text)
- MeitY advisory on AI platforms and religious content, 9 Feb 2026, accessed 2026-10-19
- MeitY draft Second Amendment Rules, 2026 (Rule 3(4)), accessed 2026-10-19
- Secondary, used only to fill the gap on March 2024 wording: AZB and Partners update, 21 Mar 2024; PSA Legal analysis; Ahlawat and Associates note, accessed 2026-10-19
This is general information, not legal advice. Last verified 19 Oct 2026 against the official source. Laws change; confirm the current position with qualified counsel before acting.