India AI & sector rulesUpdated 22 November 20268 min read

AI Labelling in India vs EU AI Act Article 50

Side-by-side comparison of India's IT Rules synthetic content labelling and EU AI Act Article 50 transparency duties, with a single build plan that can satisfy both.

By Suhas Bhairav · Reviewed against the DPDP Rules, 2025 (G.S.R. 846(E), 13 Nov 2025)

India instrument
IT (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021, as amended Feb 2026
EU instrument
Regulation (EU) 2024/1689, Article 50
EU Article 50 applies from
2 August 2026
India draft published
22 October 2025
EU max fine (most obligations)
€15m or 3% of worldwide turnover
India enforcement lever
Loss of safe harbour under s.79 IT Act

Quick answer

India and the EU both require AI-generated content to be identifiable, but through different laws. India uses IT Rules, 2021 amendments (draft 22 October 2025, final notified February 2026) aimed at intermediaries, with visible labels and metadata. The EU AI Act's Article 50, applying from 2 August 2026, puts machine-readable marking on providers and disclosure duties on deployers.

  • India's labelling duty sits in intermediary rules; the EU's sits in an AI product law with separate provider and deployer duties.
  • EU Article 50 requires machine-readable, detectable marking of synthetic audio, image, video and text from generative systems.
  • India's October 2025 draft focused on visible labels plus metadata; the February 2026 final text changed draft specifics, so check it.
  • Article 50 also requires chatbots to tell people they are talking to AI unless obvious, which India's IT Rules do not directly address.
  • A single pipeline with visible labels, embedded provenance, watermarking and chatbot disclosure can meet both regimes.

Who this is for: GenAI product teams selling in India and the EU, Trust-and-safety and content policy leads, Chatbot and voice assistant builders, Media, marketing and publishing companies, Compliance teams at Indian SaaS exporters

How do India's and the EU's AI labelling rules differ?

Both regimes aim to stop people being deceived by synthetic content. They differ in who is bound, what must be marked and how it is enforced.

India regulates through the IT Rules, 2021, which bind intermediaries. MeitY's draft of 22 October 2025 introduced "synthetically generated information" and proposed visible labels and metadata; the final amendment was notified in February 2026 with changes to draft specifics.

The EU regulates through Article 50 of the AI Act (Regulation (EU) 2024/1689), applying from 2 August 2026. It assigns duties to providers (who build or place AI systems on the market) and deployers (who use them), and it covers chatbots as well as generated media.

If you run a GenAI product used in both markets, you need to satisfy both. Fortunately, the controls overlap: visible labelling, machine-readable provenance and disclosure of AI interaction.

Side-by-side comparison

DimensionIndia (IT Rules amendments)EU AI Act Article 50
Legal basisIntermediary due diligence under the IT ActAI product regulation
Primary addresseeIntermediaries, especially significant social media intermediariesProviders and deployers of AI systems
Chatbot disclosureNot a specific IT Rules dutyProviders must design systems so people know they are interacting with AI, unless obvious
Generated mediaVisible label and metadata (draft: ≥10% visual area / first 10% of audio; check final)Providers mark outputs in machine-readable, detectable form
TextCovered as information generally; check final scopeGenerated text marked by providers; deployers disclose AI text on public-interest matters unless human-reviewed with editorial responsibility
DeepfakesLabelling and user declarationsDeployers must disclose that content is artificially generated or manipulated
Emotion recognition / biometric categorisationNot specifically addressedDeployers must inform exposed people
EnforcementLoss of s.79 safe harbour; IT Act actionFines up to €15m or 3% of worldwide turnover
Supporting guidanceCheck MeitY advisoriesCommission-facilitated code of practice on marking and labelling

The "check final" notes matter: the February 2026 Indian text changed draft specifics, so confirm thresholds against the Gazette before building.

India's approach in more detail

India's rules are platform-centric. The concern is synthetic content spreading through intermediaries, so the obligations cluster at upload and distribution.

What the October 2025 draft proposed

  • a definition of synthetically generated information;
  • visible labels and embedded metadata or identifiers;
  • label prominence of at least 10% of visual area, or the first 10% of audio duration;
  • for significant social media intermediaries: user declarations on whether uploads are synthetic, verified with reasonable technical measures.

What to assume about the final text

Final amendments were notified in February 2026. The final text changed several of the draft's specifics. Treat the draft figures as historical context only. Pull the notified amendment from egazette.gov.in and record the exact obligations in your compliance register.

Underlying duties

Rule 3(1)(b) due diligence, takedown on lawful orders and grievance redressal continue to apply. Non-compliance risks loss of safe harbour under section 79. For the full Indian picture, see India deepfake and synthetic content rules.

The EU's Article 50 in more detail

Article 50 creates five transparency duties, split by role.

Provider duties

  1. AI interaction. Systems intended to interact directly with people must be designed so those people are informed they are interacting with an AI system, unless that is obvious from context.
  2. Machine-readable marking. Providers of systems generating synthetic audio, image, video or text must ensure outputs are marked in a machine-readable format and detectable as artificially generated or manipulated.

Deployer duties

  1. Emotion recognition and biometric categorisation. Deployers must inform the people exposed.
  2. Deepfakes. Deployers must disclose that the content has been artificially generated or manipulated.
  3. Public-interest text. Deployers publishing AI-generated text to inform the public on matters of public interest must disclose it, unless the text has undergone human review and someone holds editorial responsibility.

Territorial reach

Article 2 applies the AI Act to providers placing systems on the EU market wherever they are established, and to providers and deployers outside the EU where output is used in the EU. An Indian startup with EU users is in scope. The Article 50 checklist for Indian GenAI startups turns this into tasks.

The Commission is facilitating a code of practice on marking and labelling; check its status on the EU AI Act overview page.

One build plan that satisfies both

Running two labelling systems is wasteful. Design one pipeline with configurable parameters per market.

ControlSatisfies IndiaSatisfies EU
Visible label on generated images, video, audioYes (core duty)Supports deepfake disclosure by deployers
Embedded provenance metadataYes (metadata)Yes (machine-readable marking)
Robust invisible watermarkSupports detectionSupports "detectable" requirement
Text provenance (metadata or statistical watermark where feasible)Check final scopeYes for generated text
"You are chatting with an AI" bannerGood practiceYes (provider duty)
Upload declaration and detectionYes (SSMIs)Helps deployers meet deepfake duty
Label configuration by regionMatches final Indian specificsMatches code of practice

Engineering notes

  • Apply marking at generation, server-side, so it cannot be skipped by API users.
  • Preserve metadata through resizing and transcoding; test it in CI.
  • Log label decisions. Under DPDP, those logs contain personal data and must be retained for at least one year under Rule 8(3).
  • Expose a detection endpoint or documentation so downstream deployers can check your marks.

Neither labelling regime handles the privacy of the people in the content. In India, that is the DPDP Act's job from 13 May 2027.

  • Labelling a deepfake as synthetic does not make it lawful to process the depicted person's personal data. You still need consent or a section 7 legitimate use.
  • A chatbot disclosure under Article 50 is not a DPDP notice. Rule 3 requires an itemised description of personal data, the specified purpose and links to withdraw consent, exercise rights and complain to the Board. See DPDP privacy notice requirements.
  • For EU users, GDPR applies in parallel to the AI Act; see selling AI SaaS to Europe from India.

Common mistakes

  • Treating an AI disclaimer in the footer as Article 50 compliance.
  • Marking only images and forgetting generated audio and text.
  • Hard-coding India's draft 10% figure.
  • Assuming labelling removes the need for consent from people depicted.

Further reading on suhasbhairav.com

Before 13 May 2027

Compliance checklist

  1. List every generative output type

    Inventory image, video, audio and text outputs per product and market.

    AI Act Art. 50(2)

  2. Add AI interaction disclosure

    Show a clear AI disclosure at the start of chatbot and voice sessions.

    AI Act Art. 50(1)

  3. Mark outputs at generation

    Embed machine-readable provenance and a watermark server-side for every synthetic output.

    AI Act Art. 50(2); IT Rules amendments

  4. Configure visible labels

    Apply visible labels with size and placement driven by per-market configuration.

    IT Rules amendments (Feb 2026)

  5. Check the notified Indian text

    Confirm final label specifics and timelines from the February 2026 Gazette notification.

    IT Rules, 2021 (as amended)

  6. Brief deployers

    Document how enterprise customers can meet their deepfake and public-interest text disclosure duties.

    AI Act Art. 50(4)

  7. Separate DPDP notices

    Keep Rule 3 privacy notices distinct from AI disclosure banners.

    DPDP Rule 3

Key dates

  1. EU AI Act enters into force

  2. MeitY publishes draft synthetic content amendments to IT Rules

  3. EU AI Act Article 50 transparency obligations apply

  4. DPDP core obligations apply in India

Answers

Frequently asked questions

Does the EU AI Act require watermarking AI images?

Article 50 requires providers of generative AI systems to mark synthetic audio, image, video and text outputs in a machine-readable format so they are detectable as AI-generated. It does not name one technology; metadata, watermarks or both can be used. A Commission-facilitated code of practice addresses marking techniques.

Does India require AI-generated content labels?

Yes, through the IT Rules, 2021 as amended. MeitY's October 2025 draft required visible labels and metadata on synthetically generated information, and final amendments were notified in February 2026 with changes to draft specifics. Intermediaries are the main addressees; check the notified Gazette text.

Is a chatbot disclosure required in India?

India's IT Rules amendments focus on synthetic content rather than chatbot interaction. The EU's Article 50 does require chatbots to inform users they are interacting with AI unless obvious. RBI's FREE-AI report also recommends AI disclosure for banks. Adding a disclosure is low-cost and good practice in India.

When does Article 50 apply?

Article 50 applies from 2 August 2026 under the AI Act as adopted. The Commission's November 2025 Digital Omnibus proposal would change some AI Act timelines; check its adoption status on EUR-Lex and the Commission's AI Act page before relying on any revised date.

Can one labelling system satisfy both India and the EU?

Largely yes. Combine visible labels, embedded provenance metadata, robust watermarking and AI interaction disclosure, with per-market configuration for label size and placement. Then confirm the Indian final text and the EU code of practice for any specific formats.

Do labelling rules replace privacy compliance?

No. Labels address deception, not privacy. If synthetic content depicts a real person, India's DPDP Act requires a lawful ground, security safeguards and erasure on withdrawal from 13 May 2027, and GDPR applies to EU individuals.

Legal references and sources

  • EU AI Act, Regulation (EU) 2024/1689 — Article 50 (official text)
  • IT (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021 — Synthetic content amendments (Feb 2026); Rule 3(1)(b) (official text)
  • Digital Personal Data Protection Rules, 2025 — Rule 3 (notice) (official text)

This guide explains the law for engineering and business teams. It is not legal advice; confirm how it applies to your organisation with qualified counsel, and check the Gazette for later notifications or amendments.

About the author

Suhas Bhairav

Suhas Bhairav

Systems architect and applied AI practitioner who designs production AI systems for Indian and global companies, with a focus on privacy-by-design under the DPDP Act and the EU AI Act.