Mandatory labelling of AI-generated content: rules coming soon
The government has wrapped up industry consultations on mandatory labelling for AI-generated and AI-modified content. According to IT Secretary S Krishnan, the rules are now in inter-ministerial review and will be issued shortly.
Industry feedback has been constructive. The key ask: clear guidance on what counts as a substantive AI change versus routine technical enhancements that improve quality without altering facts.
What the draft rules point to
- Clear labelling of AI-generated or materially AI-modified content with visible markers.
- Prominence: labels covering at least 10% of the visual display, or the first 10% of an audio clip's duration.
- Metadata embedding for traceability and verification across platforms.
- Greater accountability for large platforms to verify and flag synthetic media to curb deepfakes and misinformation.
Why this matters for government teams
Deepfakes and synthetic media can mislead the public, damage reputations, skew elections, and enable fraud. Citizens have the right to know if a piece of content is synthetic or authentic. Even small AI edits can flip meaning-changing one or two words in a quote or altering a voice sample can shift public perception.
Substantive AI changes vs routine enhancements
Expect the rules to differentiate between material edits (label) and minor improvements (likely don't label). Use this practical split when drafting internal SOPs:
- Label (material change): AI-created images or video; face/voice swaps; altered quotes or captions; AI-written summaries that change tone or meaning; synthetic background replacement that changes context; cloned voices; AI-edited transcripts that modify statements.
- Likely no label (routine enhancements): Exposure/contrast fixes; noise reduction; color correction; de-noising wind in audio; stabilization; minor cropping or compression-so long as facts and meaning remain intact.
When in doubt, label. Small edits can carry big consequences.
Action checklist for departments, agencies, and PSUs
- Audit your workflows: Map where AI is used across content creation, editing, and distribution. Keep logs of tools, prompts, and outputs.
- Define "label triggers": Write a one-page policy that names edits requiring a label. Include examples and edge cases.
- Standardize labels: Create overlays that meet the "10% of display" rule. For audio, add an opening spoken disclaimer or tone plus a brief verbal label in the first seconds.
- Embed metadata: Add provenance tags to files at source and maintain them through the content pipeline. Verify that common platforms don't strip this data.
- Update contracts and RFPs: Require vendors and media partners to label, embed metadata, keep process logs, and support takedowns or corrections.
- Incident response: Set up a rapid verification channel for suspected deepfakes, a correction playbook, and evidence preservation steps.
- Training: Run short sessions on responsible AI use, deepfake risks, and your labelling SOP. Keep a simple FAQ for staff.
- Governance: Assign an approving authority for edge cases and a periodic review of sampled content for compliance.
Implementation notes
- Use consistent wording and color for labels. Provide labels in the local language of the audience when needed.
- Place labels where they survive cropping and reshares. Test on mobile, desktop, and TV formats.
- For multi-clip videos, label at the start and before each synthetic segment if segments are separated.
- Keep an audit trail: original files, AI prompts/parameters, tool versions, and approval decisions.
What's next
The IT Ministry is finalizing changes after internal review. Departments should prepare now so compliance is smooth once the notification drops. Monitor updates from the Ministry of Electronics & IT for the final text and timelines.
Ministry of Electronics & IT (MeitY)
Need a quick skills refresh for your team?
For practical upskilling on AI literacy, prompts, and oversight roles, see curated options by job function.
AI courses by job role - Complete AI Training
Bottom line: labelling is coming. Get your policy, templates, metadata, and training in place now, so your next release is compliant on day one.
Your membership also unlocks:
