Relevance (UPSC): GS-II Governance (Cyber law, Platforms) | GS-III Science & Tech (Artificial Intelligence, Deepfakes)

A draft change to the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021 proposes a simple idea with big consequences: every synthetic or Artificial Intelligence-generated post on social platforms must carry a clear label. If a user does not declare it, the platform must detect and label it proactively. The push comes amid rising deepfake harms—impersonation of public figures, fraud, and doctored images that damage reputations and public trust.

What the draft says

• Coverage: All synthetic content—video, photo, audio and text—must carry a disclosure.
  
• Visibility rule: For audio-visual posts, the disclosure should cover at least ten percent of the visual surface area; for audio, a spoken or audible disclosure at the beginning (and on re-share) is expected.
  
• Self-declaration first: Platforms must provide an easy “this is AI-generated” toggle to users.
  
• Platform duty next: When users fail to declare, platforms must identify and label such content.
  
• Scope: The rule applies beyond photorealistic deepfakes—cartoons, stylised images and synthetic text are included.
  
• Enforcement lever: Non-compliance can invite directions under the Information Technology Act and threaten an intermediary’s “safe harbour” (legal shield) if due diligence is not followed.
  

How this fits into India’s digital rule-book

• Information Technology Rules, 2021: already impose due-diligence on intermediaries to act on unlawful content and to follow additional obligations for “significant” social-media platforms. The new draft adds a transparency duty for synthetic media.
  
• Digital Personal Data Protection Act, 2023: strengthens consent and fairness in data use—relevant when faces, voices or biographies are scraped to create deepfakes.
  
• Criminal law and blocking powers: impersonation, cheating and obscenity may attract action under criminal law; Section 69A blocking directions can be used for harmful deepfakes.
  

Why India is doing this

• Protect citizens: elderly people, students and migrants are being trapped by impersonation scams and fake “video calls” with forged court orders or police threats.
  
• Protect public discourse: doctored speeches and images spread quickly during elections or crises, eroding trust in institutions.
  
• Give users a fair choice: a visible tag helps people decide whether to believe, share or report content.
  

What counts as “good labelling”

• Clear, persistent and language-appropriate badges on the post and the preview, not hidden in menus.
  
• Audible disclosure for audio; text-to-speech for visually impaired users.
  
• Metadata watermarks (for machine checks) plus on-screen labels (for humans). Global standards such as C2PA style content credentials can help.
  

Implementation challenges—and practical fixes

1. Accuracy vs. over-blocking
   
   • Risk: False labels may chill satire or art; missed labels let harms slip through.
     
   • Fix: Combine user self-declarations, cryptographic watermarks output by creation tools, and post-upload detection. Allow appeal and correction within tight timelines.
     
2. End-to-end encrypted services
   
   • Risk: Harder to scan private messages.
     
   • Fix: Focus obligations on public posts; promote creator-side watermarks and receiver-side education for private shares.
     
3. Local languages and accessibility
   
   • Fix: Notify a standard label icon and short phrases in major Indian languages; require screen-reader compatibility.
     
4. Cross-border platforms
   
   • Fix: Tie compliance to safe-harbour eligibility and grievance timelines; coordinate with Election Commission of India for poll-time advisories.
     
5. Misuse of labels
   
   • Fix: Penalise false “AI” tagging used to dodge defamation or mislead; publish transparency reports on detection and appeals.
     

How India compares globally

• The European Union Artificial Intelligence Act mandates transparency for deepfakes with limited exceptions (art, research, law enforcement).
  
• The United States is pursuing voluntary watermarks and provenance standards with large model providers.
  
• India’s proposal goes further by specifying label size (ten percent) and placing a duty to detect on platforms—useful for a populous, multilingual environment.
  

What this means for citizens, creators and platforms

• Citizens: learn to look for the label, slow down before sharing, and report unlabeled suspicious posts.
  
• Creators and influencers: declare synthetic work; keep behind-the-scenes files to prove provenance; avoid misleading edits in ads and political messaging.
  
• Platforms: build easy labelling flows, automated and human review, local-language badges, and public dashboards showing the share of labelled synthetic content and action taken.
  

Key terms

• Synthetic content / deepfake: media created or altered by algorithms to look or sound real.
  
• Intermediary: a platform that hosts or transmits user content (social networks, video-sharing, messaging services).
  
• Safe harbour: legal protection for intermediaries if they follow due-diligence rules.
  
• Watermark / provenance: hidden or visible markers that show where and how a file was made.
  
• Proactive detection: platform-side systems that identify and tag likely synthetic posts without waiting for complaints.
  

Exam hook

Use this topic to connect technology harms with regulatory design: show how labelling + detection + user choice + due process can reduce deepfake risks while protecting speech and innovation. Cite the Information Technology Rules, 2021, the Data Protection Act, 2023, and election-time safeguards.

Key takeaways

• India plans mandatory labels for all AI-generated content, with a ten percent surface-area rule for visuals and platform duty to detect.
  
• The approach protects citizens from impersonation and fraud and preserves trust in public discourse.

• Success needs clear standards, multilingual labels, creator-side watermarks, appeals, and regular transparency reporting.
  

Using in the Mains Exam

Structure answers as Context → What the draft mandates → Why needed → Implementation challenges and fixes → Legal hooks (IT Rules, Data Protection) → Global comparison → Way forward (standards, transparency, citizen literacy).

UPSC Mains question

“Transparency, not takedown, is the first line of defence against deepfakes.” Examine India’s proposal to mandate labelling of AI-generated content on social media. Discuss benefits, risks to free expression, enforcement under the Information Technology Rules, 2021, and safeguards you would add.

UPSC Prelims question

With reference to India’s proposed rules on synthetic content, consider:

1. Platforms must offer users a way to self-declare that a post is AI-generated.
   
2. A visual disclosure covering at least ten percent of the post is envisaged for audio-visual content.
   
3. Non-compliance risks loss of safe-harbour protections available to intermediaries.
   Which are correct? 1, 2 and 3.
   

One-line wrap: Tell people when a post is machine-made—clear labels, smart detection and fair appeals can keep creativity alive and deepfakes in check.