Deepfakes, AI-Generated Misinformation: Technology, Regulation, and Society
June 1, 20267 min read
The question reads: "With reference to deepfake technology, consider the following statements:"
A video of an actor saying something she never said. A voice clone of a politician making a speech she never made. A photograph that never existed of a person who never was. Deepfakes — synthetic media created or manipulated by artificial intelligence — have moved from research labs to the mainstream. The technology is not new (face-swapping apps existed in 2017), but the scale, quality, and accessibility have changed dramatically with generative AI. UPSC will test: what are deepfakes, how do they threaten democracy, what is India's regulatory framework, and what are the proposed solutions?
Primary GS paper: GS-3 (Science and Technology — AI applications and risks)
[EXAMINER REASONING]
Primary trap. Candidates confuse "deepfakes" with all forms of misinformation. Deepfakes are a specific subset of AI-generated or AI-manipulated media — synthetic images, videos, or audio created by generative AI models (GANs — Generative Adversarial Networks, diffusion models, or voice cloning models). Misinformation (false information spread regardless of intent), disinformation (deliberately false), and malinformation (truthful information shared with malicious intent) are broader categories. A deepfake is always AI-generated or AI-manipulated; not all misinformation is deepfake.
Most confused. The legal position: India does NOT have a specific law against deepfakes. The existing legal framework uses: Section 66D of the IT Act (punishment for cheating by impersonation using computer resource — up to 3 years imprisonment, fine up to ₹1 lakh), Section 66E (privacy violation — publication of private images), Section 67/67A (obscene content), and IT Rules 2021 (intermediary obligation to remove "false" content flagged by government or affected individuals). None of these explicitly address AI-generated content — they use analogies to existing offences. The proposed Digital India Act (DIA) is expected to include specific deepfake provisions, but it has not been enacted.
Key anchor. The distinction between "restricted" and "generative" deepfakes. Restricted deepfakes target specific individuals (political leaders, celebrities, private citizens) — regulated through existing identity theft and defamation laws. Generative deepfakes create entirely synthetic content (AI-generated news anchors, fictional events) — these fall into a regulatory gap because there is no "victim" who is being impersonated. The "no victim" problem is the central regulatory challenge: who complains to the platform? Who has standing to seek removal?
Current affairs hook. The 2024 Indian general election was described as the "first AI election" — an estimated 50-100 million Indians viewed AI-generated content during the campaign. Deepfake videos of Bollywood actors endorsing candidates, AI-generated voice clones of party leaders making speeches (including a controversial deepfake of Rahul Gandhi allegedly conceding defeat before results), and AI-generated images of candidates performing relief work. The Election Commission issued advisories but had no legal power to mandate takedown. Platforms (Facebook, YouTube, X, WhatsApp) relied on voluntary labelling policies.
Mains hinge. The tension between free expression and regulation. Deepfake regulation involves content moderation — and content moderation in India is governed by the IT Act Section 79 safe harbour (intermediaries are not liable for third-party content if they respond to takedown orders) and IT Rules 2021 (traceability of first originator, grievance redressal officer, transparency reporting). The challenge: deepfake detection technology is imperfect, automated takedown risks over-censorship, and the traceability requirement (knowing who uploaded a deepfake) creates privacy concerns. The policy answer must balance three goals: (a) removal of harmful deepfakes, (b) preservation of free expression, and (c) protection of user privacy.
Core Concept
What are Deepfakes?
Deepfakes are synthetic media (video, audio, or images) in which a person's face, voice, or body is digitally altered using artificial intelligence to create realistic but fabricated content. The term combines "deep learning" (the AI technique) and "fake."
2021-2023 (Diffusion models): DALL-E, Midjourney, Stable Diffusion — text-to-image generation from natural language prompts. Democratised creation — anyone could generate synthetic images.
2023-2026 (Voice cloning + real-time video): ElevenLabs voice cloning (can mimic any voice from 30 seconds of audio), FaceSwap real-time video deepfakes (Zoom calls, live streams). Detection is increasingly difficult.
2025-2026 (Multi-modal generative AI): Sora (OpenAI), VideoPoet (Google) — realistic video generation from text prompts. Full synthetic video without source footage. The "no original" generation makes detection nearly impossible.
Applications (good and bad):
Positive uses
Negative uses
Creative industries — film post-production, dubbing, visual effects
Impersonation — celebrity endorsements without consent
Art — synthetic art, virtual influencers
Social manipulation — propaganda, communal polarisation
The Threat to Democracy:
India faces a unique deepfake vulnerability because:
Multi-lingual society: Deepfakes in any of 22 official languages can be targeted at specific linguistic communities
High WhatsApp penetration: End-to-end encrypted platforms make detection and traceability difficult
Low media literacy: India ranks low on media literacy indices — citizens are less able to identify synthetic content
Communal fault lines: Deepfakes targeting religious or caste identities can trigger real-world violence
The 2024 election saw:
AI-generated campaign songs (BJP and Congress both used AI-generated songs)
Deepfake video of actress Rashmika Mandanna (2023 case that brought deepfakes to national attention — unrelated to elections but highlighted the technology's accessibility)
Voice clone of Stalin (Tamil Nadu CM) appearing to make policy announcements
AI-generated images of Modi in military uniform (camouflage) — unauthorised use of PM's image
Multiple deepfake videos of Bollywood actors (Aamir Khan, Ranveer Singh, Ajay Devgn) endorsing candidates — leading to actors filing police complaints
Regulatory Framework in India:
Legal instrument
What it covers
Limitations for deepfakes
IT Act Section 66D
Cheating by impersonation using computer resource
Requires evidence of "cheating" — not all deepfakes are used to cheat
IT Act Section 66E
Publishing private images without consent
Deepfakes are synthetic — no "private image" exists
IT Act Section 67
Obscene content publishing
Most political deepfakes are not obscene
IT Rules 2021 (Section 3(1)(b)(v))
Intermediaries must remove content impersonating another person
Deepfakes impersonate but proving impersonation requires technical expertise
IT Rules 2021 (Section 4)
Traceability of first originator
Tech companies oppose as privacy-violating; SC challenge pending
Applies post-hoc — does not prevent real-time harm during elections
Right to Privacy (Article 21)
Puttaswamy framework
Applies to state action; private platform enforcement is voluntary
The proposed framework:
The Digital India Act (DIA) — expected to replace the IT Act 2000 — is expected to include:
Specific definition of "synthetic media" and "deepfakes"
Mandatory labelling of AI-generated content on platforms (with user-base threshold)
Criminal penalties for creation of harmful deepfakes (impersonation, fraud, electoral manipulation)
Platform obligation to deploy deepfake detection technology
Safe harbour removal: platforms lose intermediary liability protection if they fail to remove flagged deepfakes
The DIA has been under discussion since 2023. As of June 2026, the DIA Bill has been introduced in Parliament (early 2026) and referred to the Standing Committee on Communications and Information Technology. The committee is expected to submit its report in late 2026.
Global approaches to deepfake regulation:
Country
Approach
Key features
EU
AI Act (2024)
Risk-based: deepfakes classified as "limited risk" — mandatory transparency labelling. High-risk AI systems (used in elections, law enforcement, biometrics) have additional compliance obligations.
USA
No federal law; state laws (CA, TX, NY) + Executive Order (2023)
California: ban on deceptive AI-generated content in political ads within 60 days of election. Texas/New York: criminal penalties for non-consensual deepfake pornography. Federal: No comprehensive regulation — bills pending.
China
Deep Synthesis Rules (2023)
Mandatory labelling of all AI-generated content. Ban on deepfakes without consent. Registration of deep synthesis service providers. Most comprehensive regulation globally.
UK
Safety Online Act (2023)
Platforms must remove illegal content — includes deepfake intimate images. No specific deepfake law for political content. Voluntary labelling code for AI content.
India
No specific law
IT Act provisions + IT Rules 2021. DIA (proposed) expected to cover deepfakes. Government advisories (labelling, consent, 2024) — non-binding.
Detection and countermeasures:
Deepfake detection is an arms race — as detection improves, generation technology adapts:
Watermarking: Embedding digital watermarks in AI-generated content (C2PA standard — Coalition for Content Provenance and Authenticity)
Metadata analysis: Checking for absence or inconsistency in camera metadata
Blockchain verification: Registering authentic content on blockchain; any unregistered content is treated as suspicious
AI detection tools: Deepfake detection algorithms (run by Microsoft, Intel, Meta) — but accuracy is 65-85% for the best models (2025-26 state-of-the-art)
India's Ministry of Electronics and IT (MeitY) launched the National Deepfake Detection Network in 2025 — a consortium of IITs, IIITs, and NITs developing Indic-language deepfake detection. The consortium claims 78% accuracy across 12 Indian languages but is yet to be deployed on platforms.
Key Facts
Deepfakes: AI-generated/manipulated media using deep learning
GANs invented: 2014 (Ian Goodfellow)
First deepfake app: 2017 (FaceApp, DeepFaceLab)
India's 2024 election: "first AI election" — 50%+ increase in deepfake content
Known case: Rashmika Mandanna deepfake (2023) — brought national attention
India's legal framework: No specific law — IT Act (Sections 66D, 66E, 67) + IT Rules 2021
Proposed: Digital India Act (DIA) — Bill introduced 2026, Standing Committee review ongoing
Indian deepfake detection consortium: IITs + IIITs + NITs (2025), 78% accuracy in 12 languages
Global regulation: EU AI Act (risk-based), China Deep Synthesis Rules (labelling + registration), US (state-level laws)
Detection accuracy: 65-85% for best models (2025-26) — but continues to decline as technology improves
Social media platforms in India (2025 users): WhatsApp ~600M, YouTube ~500M, Facebook ~350M, Instagram ~350M
Previous Year Questions
Year
Stage
What was tested
2025
Prelims
Deepfakes — meaning and enabling technology
2024
Prelims
AI-generated content — regulatory approach in India
2024
Mains GS-3
"The proliferation of deepfakes poses a significant challenge to electoral integrity." Discuss the regulatory and technological measures needed to combat them."
2023
Mains GS-3
"Generative AI models have created both opportunities and risks for Indian society." Analyse in the context of disinformation.
2023
Prelims
GANs — what they are and their applications
2022
Mains GS-3
"Misinformation and disinformation on social media platforms threaten the democratic process." Examine India's regulatory approach.
Statement Elimination Guide
"India enacted a specific law against deepfakes in 2023." False. India does not have a standalone law against deepfakes. The government issued advisories in 2024 (requiring labelling of AI-generated content under IT Rules 2021) but these are executive instructions, not legislation. No specific deepfake law exists.
"The IT Act Section 66D (cheating by impersonation) applies to all deepfakes." False. Section 66D requires proof of "cheating" — that the impersonation was used to deceive someone for gain or cause harm. Many deepfakes (satirical, artistic, or those created to embarrass) do not meet the cheating threshold. Section 66D covers some deepfakes but not all.
"The Digital India Act (DIA) is expected to replace the IT Act 2000 and include specific deepfake provisions." Correct. The DIA (proposed) will repeal/replace the IT Act 2000 and is expected to include a definition of synthetic media, mandatory labelling obligations, platform detection requirements, and criminal penalties for harmful deepfakes. The Bill was introduced in Parliament in early 2026.
"MeitY's 2024 advisory on AI labelling is a binding legal requirement." False. The advisory initially (March 2024) required government approval before deploying AI models. After industry backlash (including Google, Microsoft, and Indian AI startups threatening to leave), the advisory was withdrawn and replaced with a labelling requirement. Even the labelling requirement is an advisory — not backed by statutory penalties for non-compliance.
"China has the most comprehensive deepfake regulation globally." Correct. China's Deep Synthesis Rules (2023) require all AI-generated content to be labelled, service providers to register with authorities, and users to consent to deepfake creation. Penalties include business licence revocation and criminal liability. China also has the world's strictest social media content moderation, making deepfake enforcement more feasible than in democratic countries.
Current Affairs Hook
The 2026 Census and the upcoming 2029 general election create urgency for deepfake regulation. MeitY convened an inter-ministerial group in early 2026 to draft deepfake-specific provisions that could be added to the DIA as the Bill proceeds through Parliament. The proposed provisions include: mandatory AI labelling for all synthetic media exceeding 10,000 views, creation of a deepfake rapid response mechanism within the Election Commission, and a "right to know" for citizens — platforms must disclose whether content they view is AI-generated.
The 2025 Supreme Court case on deepfakes (filed by NGO Software Freedom Law Centre) is pending. The petition argues that the absence of a specific legal framework violates Article 19(1)(a) (right to information — citizens cannot trust information to form public opinion) and Article 21 (right to reputation — victims of non-consensual deepfakes have no effective remedy). The government has countered that existing IT Act provisions are adequate and that the DIA will address remaining gaps. The case outcome will shape deepfake regulation.
The Election Commission issued a consultation paper in December 2025 proposing new rules: (1) mandatory pre-certification of AI-generated campaign content, (2) 72-hour takedown deadline for flagged deepfakes during election periods, (3) transparency reporting by platforms on AI-generated content removal. Political parties have opposed these rules, arguing they would restrict legitimate campaign speech (including AI-generated campaign songs, synthetic messages in multiple languages).
Interlinkages
Science & Technology: Deepfakes are an application of generative AI — specifically GANs (generative adversarial networks) and diffusion models. The same technology that powers creative tools (DALL-E, Midjourney) powers deepfakes. India's National AI Mission (₹10,372 crore) includes a "safe and trusted AI" component but has allocated only ₹100 crore to deepfake research — critics argue this is insufficient given the scale of the threat.
Internal Security: Deepfakes can be used by state-sponsored actors for disinformation warfare — targeting Indian elections, spreading communal polarisation, undermining public institutions. CERT-In's cyber threat assessment (2025) identified AI-generated disinformation as a "top 3 emerging threat" along with ransomware and state-sponsored APTs.
Governance: The intermediary liability framework (IT Act Section 79, IT Rules 2021) governs platform responsibility for deepfake removal. The "safe harbour" protection — platforms are not liable for user-generated content if they respond to takedown orders — creates an incentive structure: remove flagged content quickly, but no obligation to proactively detect. Deepfake detection is proactive, which the current regulatory framework does not require.
Polity: Deepfakes test the boundaries of free expression (Article 19(1)(a)). The reasonable restrictions under Article 19(2) (security of state, public order, defamation, incitement to offence) provide the constitutional basis for deepfake regulation — but the "proportionality" requirement from the Puttaswamy framework means deepfake regulation must be narrowly tailored.
Ethics: India's AI ethics framework — NITI Aayog's Responsible AI papers (2020-2023) — identifies "transparency and explainability" as core principles. Deepfakes directly violate these principles: they are designed to deceive about their origin. The ethics framework recommends mandatory disclosure of AI-generated content but has no enforcement mechanism.
Common Mistakes
Thinking deepfakes require high technical expertise. Since 2023, voice cloning requires just 30 seconds of audio. Face-swap apps run on smartphones. AI-generated video (Sora, 2024+) creates realistic footage from text prompts. The barrier to entry is near zero.
Assuming all deepfakes are illegal. Many deepfakes are legal — satirical deepfakes, artistic works, parody, and education fall under free expression. The illegality depends on the purpose (impersonation for cheating, fraud, defamation, electoral manipulation) — not the technology itself.
Confusing platform labelling policies with legal requirements. Major platforms (Meta, Google, X, TikTok) introduced voluntary AI labelling in 2024-25. These are company policies, not legal mandates. The government's 2024 advisory requires labelling under IT Rules — but since the Rules lack explicit deepfake provisions, the legal basis is contested.
Believing deepfake detection can fully solve the problem. Detection technology is an arms race: as detection improves, generation adapts. Current best detection accuracy is 65-85% — and degrades when content is compressed, re-encoded, or distributed on low-resolution platforms (WhatsApp, where videos are compressed). There is no "technical solution" — regulation, media literacy, and detection must work together.
Overlooking India's "right to be forgotten" connection. Victims of non-consensual deepfake pornography can seek removal under the DPDP Act 2023 (right to erasure of personal data) — but the DPDP Act requires the content to qualify as "personal data," which deepfakes (synthetic content about a person) may not. The interaction between deepfakes and data protection law is legally uncertain.
Revision Snapshot
Deepfakes = AI-generated/manipulated synthetic media. Technology evolution: GANs (2017) → Diffusion models (2022) → Real-time voice/video cloning (2024) → Full synthetic video generation (2025-26). Indian regulation: NO specific deepfake law — IT Act Sections 66D (impersonation for cheating), 66E (privacy), 67 (obscenity) applied by analogy. 2024 IT Rules advisory: mandatory labelling of AI-generated content (non-binding). Proposed DIA (Digital India Act, Bill introduced 2026) expected to cover deepfakes — definition, labelling, detection obligations, penalties. Global models: EU AI Act (risk-based labelling), China Deep Synthesis Rules (registration + labelling), US (state laws). Detection accuracy: 65-85% (2025-26). India's vulnerability: multilingual population, encrypted platforms (WhatsApp), low media literacy. UPSC takeaway: deepfakes test the intersection of technology, free speech, election integrity, and regulation — India has no specific law yet, the DIA is the pending solution.