India Tightens Labels for AI-Generated Content

India's Ministry of Electronics and IT (MeitY) circulated draft amendments to the IT Rules that replace the existing "prominent visibility" requirement with a mandate for continuous on-screen labels on AI-generated and AI-manipulated content for the full duration of playback. The draft extends to fully synthetic material and content "significantly altered" by AI, and it requires intermediaries to prevent removal or suppression of labels while embedding provenance markers; the public consultation deadline was extended to May 7, 2026.

The proposal operationalizes labeling in two technical layers. First, a visible overlay or watermark that must remain on-screen throughout playback, so viewers cannot miss the disclosure if they join midstream. Second, embedded metadata or provenance markers that travel with the file or stream and cannot be stripped by intermediaries. Practitioners should note these design and implementation implications:

• •Persistent visual markers that can withstand re-encoding, cropping, and player overlays
• •Embedded metadata and provenance that survive CDN caching and third-party reposting
• •Requirements for audio markers or disclosures for voice-cloned content and altered audio

Implementing continuous overlays raises technical tradeoffs: label placement, size, opacity, and screen-area minimums (some reports reference proposals for minimum coverage) affect accessibility and UX. Resilient metadata requires standardized fields and cryptographic signing to avoid easy removal or spoofing. Detection tooling must distinguish between innocuous edits and "significant" AI manipulation, which will drive classifier design and false-positive management.

This amendment continues India's trend of treating synthetic media as a distinct regulatory object, building on the February 2026 IT Rules changes that tightened labeling and takedown expectations. For platforms and ML teams, the rule closes a loophole where short or early disclosures were effectively invisible to downstream viewers. It also accelerates demand for practical watermarking and provenance solutions from both open-source projects and vendor ecosystems. The law intersects with several ongoing industry trends: the emergence of robust watermarking libraries, provenance standards like content credentialing, and platform engineering to enforce UI-level constraints across web, native apps, and third-party embeds.

Continuous labels are technically straightforward for native uploads but much harder for live streams, third-party embeds, and rehosted content. Adversaries can attempt to crop, overlay, or re-encode to remove visual markers; without cryptographic provenance, labels are brittle. Platforms will need to balance label visibility with accessibility, minimize disruption to UX, and update moderation pipelines to treat label removal as an enforcement trigger. For ML teams, the rule increases demand for robust detection models, watermarking at model-training time, and interoperable metadata standards.

Monitor MeitY's final text after the consultation window and any technical guidance or implementation timelines from major intermediaries. Key signals will include whether regulators specify label size or cryptographic requirements, whether standards bodies propose interoperable metadata schemas, and how platforms handle live and ephemeral content. Expect vendor offerings for robust watermarking, provenance signing, and detection-as-a-service to accelerate in response.

The draft raises the bar on transparency by making AI markers persistent and tamper-resistant in expectation. It creates immediate engineering priorities for platforms and ML teams: robust watermarking, hardened metadata provenance, improved detection of AI manipulations, and UX approaches that preserve accessibility while keeping labels unmissable.