Authored By: Aman Gupta
Shri Ramswaroop Memorial University
Introduction
The quick rise of AI deepfakes is becoming a serious threat to fair elections.[1] In India, where political parties rely heavily on mass texts, short video apps, and intense campaigning, fake media can change voters’ minds much faster than regular rumours ever could. The real problem isn’t just false information spreading—it’s that realistic fake videos and audio clips can impersonate politicians, make up speeches, and trick the public before anyone has time to fact-check them. Recent steps by the government, Digital Personal Data Protection Act, 2023 and new warnings from the Election Commission,[2] show that officials are starting to realize how dangerous this is.
However, India’s current laws aren’t ready to handle the specific problems deepfakes cause during elections. Existing rules can deal with basic issues like defamation or identity theft, but they don’t offer a clear, focused plan for AI-generated political content. To fix this, India needs mandatory labels on AI content, fast removal of fake videos during election season, and fair penalties for the people who create, fund, or host harmful deepfakes.
This article is divided into five parts. First, it looks at India’s current laws. Second, it reviews past court cases and explains why present legal remedies aren’t enough. Third, it points out the main weaknesses in how India currently handles election deepfakes. Finally, it suggests a practical solution that protects free speech while keeping elections safe from manipulation.
Legal Framework
India does not yet have a standalone statute that comprehensively regulates deepfakes in elections.[3] Instead, legal control is dispersed across election law, criminal law, defamation principles, information technology regulation, and data protection. That fragmentation matters because deepfakes do not fit neatly into any one category of wrong. A synthetic clip may be defamatory, impersonate a candidate, misuse personal data, or mislead voters without clearly violating every element of any single offence or civil claim.
At the level of election law, the Representation of the People Act, 1951[4] remains the principal statute governing electoral conduct. It contains provisions on corrupt practices, electoral offences, and the maintenance of free and fair elections. However, these provisions were drafted in an era of print, rallies, and broadcast media, not AI-generated impersonation. Their language can reach some forms of false campaigning, but they do not specifically identify synthetic media, manipulated voice cloning, or algorithmically generated political impersonation.
The Information Technology Act, 2000[5] and the intermediary regime also provide partial tools. Platform intermediaries may be required to act against unlawful content once notified, and the law enables action against certain computer-related offences. Yet the existing framework focuses more on intermediaries as hosting channels than on synthetic content as a distinct electoral threat. The result is a reactive model that depends heavily on notice, platform discretion, and later-stage enforcement.
The Digital Personal Data Protection Act, 2023[6] is also relevant because deepfakes often rely on personal data, voice samples, face images, and other identifiable material. The Act’s stated purpose is to regulate digital personal data processing while recognising the individual’s right to protect such data and the need for lawful processing. Its structure is important because it shifts the conversation toward accountability in data use, but it does not directly answer the electoral harms caused when lawful data is repurposed into manipulated political media.
More recently, the Government’s notification of the Digital Personal Data Protection Rules, 2025[7] marked the full operationalisation of the DPDP framework, emphasising consent, transparency, security safeguards, and accountability. That development is significant because it shows a policy commitment to responsible digital governance. Still, those rules remain primarily privacy-focused; they do not create a dedicated election-time regime for synthetic speech, image cloning, or manipulated candidate endorsements.
III. Judicial Context
Indian courts have long recognised that free and fair elections are part of the constitutional structure and that voter choice must remain meaningful. The Supreme Court has repeatedly treated the voter’s right to know as a component of Article 19(1)(a), especially in the context of election transparency. In Association for Democratic Reforms,[8] the Court required disclosure of candidate information so that voters could make informed choices. In PUCL v Union of India,[9] it reaffirmed that the voter’s freedom of expression includes the right to receive relevant electoral information.
That line of authority is highly relevant to deepfakes because synthetic political media directly interferes with informed electoral choice. If voters are shown fabricated footage of a candidate making statements never actually made, the injury is not only reputational; it is democratic. The harm lies in the distortion of the informational environment that the Court has already treated as constitutionally significant. The logic of the disclosure cases therefore supports a stronger regulatory response to manipulated election content than India currently has.
The Supreme Court’s broader speech jurisprudence also matters. In Shreya Singhal v Union of India,[10] the Court struck down section 66A of the Information Technology Act, 2000[11] because vague and overbroad restrictions chilled legitimate speech. That case is a warning against any deepfake regulation that is too open-ended or too discretionary. Any election-specific framework must therefore be narrowly tailored, clearly defined, and linked to demonstrable risks such as impersonation, false attribution, and deliberate voter deception.
At the same time, the Court has recognised that speech may be restricted where other constitutional values are seriously threatened. Electoral integrity and the integrity of public discourse can justify measured constraints, especially where the speaker is not merely expressing an opinion but manufacturing a false factual record. This distinction is crucial. A democratic system must tolerate criticism, satire, and parody, but it need not tolerate machine-made falsehoods passed off as real political communication.
Why Current Law Falls Short
The main defect in the current framework is not the absence of any legal response, but the absence of a precise and speedy one. Deepfakes move at platform speed, whereas litigation, police complaints, and administrative action move slowly. By the time a false video is disputed, circulated, and later removed, the electoral harm may already be irreversible.[12] A delayed legal remedy cannot restore a damaged perception once millions of voters have already seen the content.
A second weakness is definitional. Deepfakes are technically diverse. Some are fully synthetic, some are manipulated recordings, and some are voice clones made from minimal source material. A law that uses only broad notions such as “misinformation” or “false content” may be too crude to capture the distinctive danger of synthetic political impersonation. Conversely, a law drafted too narrowly could be evaded by slight technical variation. The legal response must therefore define synthetic political media by function, not merely by production method.
A third problem concerns responsibility. Harmful deepfakes often involve multiple actors: the creator, the person who commissions the content, the political campaign that distributes it, and the platform that amplifies it. Existing law often makes this responsibility difficult to allocate in a timely and proportional manner. In practice, enforcement may focus on the most visible intermediary while leaving the strategic creator or sponsor untouched. That undermines deterrence and encourages blame-shifting.
A fourth concern is selective enforcement. Because political speech is contest-sensitive, deepfake rules can easily become tools of partisan pressure if the standards are vague or the enforcement process is opaque. This makes procedural fairness essential. Any regime must contain safeguards such as notice, reasons, an urgent review mechanism, and appeal rights.[13] Without those safeguards, regulation intended to protect democracy could itself become a threat to democratic freedom.
Comparative Lessons
Comparative experience suggests that India does not need to invent the regulatory architecture from scratch. The European Union has moved toward more structured platform accountability[14] through the Digital Services Act and emerging AI governance measures, both of which emphasise transparency, risk management, and faster response to harmful digital content. Those models are not directly transplantable, but they show that synthetic media can be regulated through layered duties rather than a single blunt prohibition.
The United States offers a different lesson. There, deepfake regulation has often taken the form of sector-specific rules, election-time disclosure obligations, and state-level interventions rather than a single federal code. That approach highlights the difficulty of regulating political media in a system that places very high value on free expression. It also shows, however, that disclosure and provenance requirements are more defensible than broad content bans.[15]
India’s own regulatory direction also points toward disclosure rather than censorship. The Election Commission’s recent advisories on AI-generated content and deepfakes suggest that the most workable path is one of transparency, traceability, and rapid correction. The current policy environment therefore supports a regime that requires political deepfakes to be labelled, traced, and removed quickly where they are deceptive or unauthorised.
Reform Model
A workable Indian framework should begin with a clear statutory definition of electoral deepfakes. The law should define them as synthetic or materially manipulated audio-visual or audio content that falsely depicts a candidate, political party, public official, or electoral event in a manner likely to mislead voters. This definition should exclude obvious parody, satire, and artistic expression that a reasonable viewer would not mistake for real political communication.
Second, the law should require prominent disclosure for all AI-generated political content used in campaigning, advertising, or voter outreach.[16] Disclosure should be visible, standardised, and unavoidable. A small footer buried in a post is not enough. If the content is synthetic, the voter should know that immediately, before the content spreads beyond the original platform. Disclosure is a less speech-restrictive measure than prohibition and is therefore easier to justify constitutionally.
Third, there should be a fast election-period takedown mechanism. During the regulated campaign window, a designated authority or electoral tribunal should be able to issue temporary removal or throttling orders within hours, subject to prompt review. The point is not to create permanent prior restraint, but to prevent irreparable harm in a time-sensitive environment where ordinary procedure is too slow. This mechanism should be limited to clearly deceptive deepfakes involving impersonation, fabricated statements, or falsified endorsements.
Fourth, liability should be tiered. The creator or sponsor of a malicious deepfake should face primary liability. The campaign that knowingly circulates it should face secondary liability. Platforms should be responsible where they ignore notice, fail to apply required labels, or do not act with due speed in election periods. This model better matches responsibility to control and knowledge than a one-size-fits-all censorship approach.
Finally, the law should require preservation of provenance data. Platforms and campaign vendors should maintain records showing when content was created, edited, uploaded, and modified. Such records would improve investigation, support remedies, and reduce the evidentiary burden on the victim. Without traceability, electoral deception remains easy and accountability remains speculative.
VII. Conclusion
Deepfakes present a unique danger to India’s democratic fabric by eroding the reliable flow of information necessary for genuine, transparent elections. While current legal mechanisms—spanning electoral codes, platform regulations, penal statutes, and data privacy frameworks—provide some recourse, they were not built to tackle AI-driven political impersonation. Consequently, the legal response remains slow, fragmented, and largely reactive against rapid digital manipulation.
Addressing this challenge requires a dedicated electoral framework centered on mandatory labeling, swift content removal, scalable platform liability, and clear digital tracing. Rather than outlawing political discourse or chilling creative parody, this targeted strategy shields citizens from synthetic disinformation while upholding constitutional protections for free expression. Striking this equilibrium is crucial to ensure democratic safeguards do not morph into tools for censorship.[17]
Ultimately, the core takeaway is straightforward: synthetic media must be addressed as a direct threat to electoral integrity, rather than a routine matter of online content management. By confronting deepfakes with appropriate legal rigor, India can safeguard both vibrant public discourse and the fundamental right of voters to make unmanipulated choices[18]—a standard vital for any constitutional democracy.
Bibliography
Cases
Association for Democratic Reforms v. Union of India, (2002) 5 S.C.C. 294.
People’s Union for Civil Liberties (PUCL) v. Union of India, (2003) 4 S.C.C. 399.
Shreya Singhal v. Union of India, (2015) 5 S.C.C. 1.
Legislation
The Digital Personal Data Protection Act, 2023, No. 22 of 2023, India Code.
Press Information Bureau, Government of India, Notification of the Digital Personal Data Protection Rules, 2025.
Information Technology Act, 2000, No. 21 of 2000.
Representation of the People Act, 1951, No. 43 of 1951.
Secondary Sources
Lilian Edwards & Michael Veale, Slave to the Algorithm? Why a Right to an Explanation Is Probably Not the Remedy You Are Looking For, 16 Duke L. & Tech. Rev. 18 (2017).
Ryan Abbott, The Reasonable Robot: Artificial Intelligence and the Law (2020).
[1] Election Commission of India, Advisory on AI-Generated Content and Deepfakes in Elections (Jan. 15, 2025).
[2] Digital Personal Data Protection Act, No. 22 of 2023 (India).
[3] Representation of the People Act, No. 43 of 1951 (India).
[4] Representation of the People Act, No. 43 of 1951 (India).
[5] Information Technology Act, No. 21 of 2000 (India).
[6] Digital Personal Data Protection Act, No. 22 of 2023 (India).
[7] Press Information Bureau, Gov’t of India, Press Release, Notification of the Digital Personal Data Protection Rules, 2025 (Nov. 19, 2025).
[8] Union of India v. Ass’n for Democratic Reforms, (2002) 5 S.C.C. 294 (India).
[9] People’s Union for Civil Liberties v. Union of India, (2003) 4 S.C.C. 399 (India).
[10] Shreya Singhal v. Union of India, (2015) 5 S.C.C. 1 (India).
[11] Information Technology Act, No. 21 of 2000 (India).
[12] Election Commission of India, Advisory on AI-Generated Content and Deepfakes in Elections (Jan. 15, 2025).
[13] Shreya Singhal v. Union of India, (2015) 5 S.C.C. 1 (India).
[14] Lilian Edwards & Michael Veale, Slave to the Algorithm? Why a Right to an Explanation Is Probably Not the Remedy You Are Looking For, 16 Duke L. & Tech. Rev. 18 (2017).
[15] Ryan Abbott, The Reasonable Robot: Artificial Intelligence and the Law (2020).
[16] Election Commission of India, Advisory on AI-Generated Content and Deepfakes in Elections (2025).
[17] Shreya Singhal v. Union of India, (2015) 5 S.C.C. 1 (India).
[18] People’s Union for Civil Liberties v. Union of India, (2003) 4 S.C.C. 399 (India).
