1. Meaning of a Deepfake

A deepfake is synthetic or manipulated audio, video, image or other media created or altered using artificial intelligence, machine learning or related computational techniques so that it appears to depict a real person, event or statement.

Examples include:

• putting a person's face onto another person's body;
• making a person appear to say something they never said;
• cloning a person's voice;
• creating fabricated intimate images;
• manipulating political speeches;
• creating fake evidence or fabricated recordings;
• impersonating a person for fraud.

The legal difficulty is that not every deepfake should be criminalised. A parody, satire, film, educational reconstruction or clearly labelled fictional work may be protected expression, whereas a fabricated sexual image used to harass someone, an impersonation used for fraud, or fabricated evidence intended to obstruct justice may justify criminal sanctions.

Therefore, the central legal question is:

What type of deepfake, created with what intention, causing what harm, should attract criminal liability?

2. Is Deepfake Creation Itself a Crime in India?

Not automatically.

India does not currently have one comprehensive offence called “deepfake” covering every form of AI-generated or manipulated content.

Instead, criminal liability may arise when the deepfake falls within an existing offence under laws such as:

• the Information Technology Act, 2000;
• the Bharatiya Nyaya Sanhita, 2023 (BNS);
• laws concerning obscenity and sexually explicit material;
• identity/personation and cheating offences;
• copyright and trademark law in appropriate cases;
• election law;
• privacy/personality-rights principles;
• child-protection law where minors are involved;
• intermediary/platform obligations.

This produces what can be called a harm-based criminalisation model.

3. The Proper Scope of Deepfake Criminalisation

A comprehensive framework should distinguish at least six categories.

This distinction is critical because an excessively broad offence could criminalise legitimate artistic and political expression.

4. Constitutional Foundation — Article 19(1)(a)

The most important constitutional limitation on deepfake criminalisation is freedom of speech and expression under Article 19(1)(a).

The State can impose reasonable restrictions under Article 19(2), including in relation to:

• sovereignty and integrity of India;
• security of the State;
• public order;
• decency or morality;
• contempt of court;
• defamation;
• incitement to an offence;
• friendly relations with foreign States.

Consequently, Parliament cannot simply criminalise:

“Any AI-generated image that is false.”

The law would need to identify a legitimate objective and satisfy constitutional requirements.

5. Shreya Singhal v. Union of India

Shreya Singhal v. Union of India, (2015) 5 SCC 1

This is one of the most important cases for regulating online speech.

The Supreme Court struck down Section 66A of the Information Technology Act because the provision was excessively vague and had a chilling effect on freedom of speech.

The Court distinguished between:

• discussion;
• advocacy; and
• incitement.

Only when speech reaches the constitutionally relevant threshold of incitement can restrictions be justified on that basis.

Relevance to deepfakes

Suppose Parliament creates an offence:

“Publishing false AI-generated content likely to cause annoyance.”

That formulation could face a serious vagueness and overbreadth challenge.

A better statute would identify specific harms, such as:

knowingly creating or distributing a materially deceptive synthetic representation of an identifiable person, with intent to defraud, sexually exploit, cause specified unlawful harm, or interfere with an electoral process.

Shreya Singhal therefore supports a narrow, precisely drafted deepfake offence rather than a blanket prohibition on synthetic media.

6. K.S. Puttaswamy v. Union of India

K.S. Puttaswamy v. Union of India, (2017) 10 SCC 1

The nine-judge Supreme Court Bench recognised privacy as a fundamental right.

This is highly relevant because deepfakes can violate:

• informational privacy;
• bodily autonomy;
• decisional autonomy;
• dignity;
• identity;
• control over personal information.

The judgment recognised informational privacy as an important aspect of privacy.

Deepfake example

A person's photograph is collected from social media and transformed into an AI-generated sexual video.

Even if the original photograph was publicly available, that does not necessarily mean:

“The person consented to any possible manipulation of their likeness.”

This is where privacy + dignity + autonomy become central.

7. R. Rajagopal v. State of Tamil Nadu

R. Rajagopal v. State of Tamil Nadu, (1994) 6 SCC 632

The Supreme Court recognised the individual's right to be let alone as an aspect of privacy.

The judgment is particularly relevant to deepfake disputes involving personal identity and unauthorised publication.

A deepfake may involve a person's:

• face;
• voice;
• body;
• name;
• reputation;
• personal attributes.

Therefore, the law needs to distinguish between legitimate public discussion and unauthorised exploitation of a person's identity.

8. Anuradha Bhasin v. Union of India

Anuradha Bhasin v. Union of India, (2020) 1 SCC 637

The Supreme Court examined restrictions on internet access and recognised the constitutional importance of freedom of speech and expression through the internet.

The broader lesson for deepfake legislation is that online speech remains constitutionally protected speech.

The government cannot treat the internet as a constitution-free zone.

Therefore:

Regulation of deepfakes must satisfy constitutional standards even when the content is distributed through social-media platforms.

9. Shreya Singhal and the Intermediary Problem

Deepfake regulation creates two separate questions:

A. Who created the deepfake?

The creator may have direct criminal liability.

B. Who hosted/distributed it?

The platform may have separate intermediary obligations.

These should not automatically be treated as identical.

For example:

Person A creates an explicit deepfake → Person B uploads it → Platform C hosts it.

The law should determine separately:

• A's liability;
• B's liability;
• C's intermediary obligations.

This is important because imposing criminal liability automatically upon platforms for every user-generated deepfake could create excessive censorship.

10. Deepfakes and the Information Technology Act

The IT Act remains relevant to deepfake conduct.

Depending on the facts, provisions concerning:

• privacy violations;
• identity theft;
• cheating by personation;
• obscene material;
• sexually explicit material;
• material involving children;

may become relevant.

Section 66C — Identity theft

Where another person's password, electronic signature or other unique identification feature is dishonestly or fraudulently used, Section 66C may become relevant.

Section 66D — Cheating by personation using computer resources

This is especially relevant to fraudulent deepfakes.

Example:

A fraudster creates an AI voice/video impersonating a company director and instructs an employee to transfer ₹50 lakh.

The deepfake is not merely “fake content.”

It is potentially an instrument of cheating/personation.

11. Deepfake Pornography

This is one of the strongest arguments for specific criminalisation.

A non-consensual sexual deepfake can cause:

• sexual harassment;
• humiliation;
• psychological harm;
• reputational damage;
• extortion;
• stalking;
• social ostracisation.

The victim may never have created an intimate photograph at all.

That is an important distinction from traditional “revenge pornography.”

Traditional situation

Actual intimate photograph → distributed without consent.

Deepfake situation

No intimate photograph may ever have existed.

AI creates a fabricated intimate depiction.

A modern criminal statute should therefore cover synthetic intimate imagery, not merely unauthorised distribution of authentic intimate photographs.

12. India: Existing Criminal Provisions Can Overlap

Depending on circumstances, deepfake pornography may engage provisions concerning:

• sexually explicit material;
• obscenity;
• voyeurism;
• harassment;
• stalking;
• criminal intimidation;
• defamation;
• offences against women;
• offences involving children.

The precise offence depends upon what was created, who was depicted, how it was distributed, the intention and resulting harm.

Therefore, one should not state that:

“Every deepfake is an offence under Section X.”

That would be legally inaccurate.

13. Child Deepfakes — Much More Serious

Where the synthetic material depicts a child in sexually explicit circumstances, the legal position becomes substantially more serious.

The Protection of Children from Sexual Offences Act, 2012 (POCSO) and the IT Act can become relevant depending on the conduct and statutory definitions.

The law should particularly distinguish between:

AI-generated sexual material depicting a fictional adult.

and

AI-generated sexual material depicting an identifiable child.

The second category raises major child-protection concerns even where the underlying image is synthetic.

14. Deepfake Fraud

Deepfakes can be used for financial crime.

Example

A criminal clones the voice of a CEO:

“Transfer ₹2 crore immediately to this account.”

The employee complies.

Here, the deepfake is merely the technology used to commit the underlying crime.

Possible legal consequences may involve:

• cheating;
• personation;
• forgery-related offences depending on the facts;
• cyber offences;
• conspiracy;
• money laundering where the broader requirements are satisfied.

The criminal law does not necessarily need a new offence for every technology.

15. Deepfake and Forgery

An important legal question is whether AI-generated material should be treated as a “false document” or equivalent evidence for forgery offences.

Under the Bharatiya Nyaya Sanhita, 2023, offences relating to forgery and use of forged documents/electronic records may become relevant depending on the nature and intended use of the synthetic material.

Example

A person creates an AI-generated video purporting to show a public official accepting a bribe and submits it as genuine evidence in litigation.

The legal analysis is very different from:

A comedian creates an obvious parody of the same official.

The intention, context, representation and legal use matter.

16. Deepfake and Evidence

This is an increasingly important area.

Deepfake technology creates a serious evidentiary problem:

How does a court determine whether audio/video evidence is authentic?

Traditionally, video evidence was often treated as comparatively persuasive because it appeared to provide a visual record.

AI undermines that assumption.

Courts may increasingly need:

• metadata analysis;
• hash verification;
• provenance records;
• forensic examination;
• source-device verification;
• chain of custody;
• expert testimony;
• comparison with original files.

The existence of deepfake technology means that digital evidence cannot automatically be equated with truth merely because it looks or sounds authentic.

17. Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal

Arjun Panditrao Khotkar v. Kailash Kushanrao Gorantyal, (2020) 7 SCC 1

The Supreme Court clarified important aspects of Section 65B of the Indian Evidence Act concerning electronic records.

Although the statutory framework has subsequently changed with the Bharatiya Sakshya Adhiniyam, 2023, the case remains highly relevant historically and conceptually.

Deepfake relevance

The case highlights the importance of properly authenticating electronic evidence.

With deepfakes becoming easier to create, the question:

“Is this recording authentic?”

becomes fundamental.

18. Deepfake and Defamation

Suppose an AI system creates a realistic video showing:

A doctor accepting a bribe.

The video is false.

The doctor's reputation may be severely damaged.

Traditional defamation principles can therefore become relevant.

However, a dedicated deepfake law should avoid simply criminalising:

“False AI content.”

Instead, it could focus on knowing or reckless fabrication presented as authentic and intended or reasonably likely to cause legally recognised harm.

This is more compatible with free-expression principles.

19. Deepfakes in Elections

Election deepfakes are particularly dangerous because they can create false impressions about candidates shortly before voting.

Examples:

• fake candidate speech;
• fabricated confession;
• fake announcement of withdrawal;
• manipulated video suggesting criminal conduct;
• fake religious or communal statements;
• fabricated government announcement.

The legal problem is balancing:

Electoral integrity

against

Freedom of political expression.

A law should therefore distinguish:

deliberate deceptive impersonation

from

political satire, criticism or parody.

20. Why Blanket Criminalisation Is Dangerous

Imagine a law saying:

“Any person who creates or distributes AI-generated content commits an offence.”

This would potentially criminalise:

• films;
• satire;
• memes;
• parody;
• educational material;
• historical reconstruction;
• artistic expression;
• political commentary.

Such a law would raise serious Article 19 concerns.

Therefore, the better approach is:

Criminalise harmful deceptive conduct, not artificial intelligence itself.

21. Mens Rea: The Most Important Element

A modern deepfake offence should specify the required mental element.

Possible standards include:

Intention

The person deliberately created the deception.

Knowledge

The person knew the material was artificially manipulated.

Recklessness

The person consciously disregarded a substantial risk of deception or harm.

Negligence

The person merely failed to take reasonable care.

For criminal liability, intention/knowledge/recklessness are generally preferable to vague negligence-based liability, particularly because of free-expression concerns.

22. Proposed Definition of a Criminal Deepfake

A future Indian statute could define a criminal deepfake approximately around these elements:

A materially manipulated or synthetically generated audio, visual or audiovisual representation that falsely depicts an identifiable person as performing, saying or appearing in conduct that did not occur, where the person knowingly creates, publishes or distributes it with specified unlawful intent or in circumstances causing specified legally recognised harm.

The definition should contain:

1. synthetic/manipulated content;
2. identifiable person;
3. material falsity;
4. representation as authentic;
5. knowledge/intention;
6. prohibited purpose or harm.

23. Possible Criminal Categories

A comprehensive statute could create separate offences.

Offence 1 — Non-consensual intimate deepfake

Creating or distributing synthetic sexual imagery of an identifiable person without consent.

Offence 2 — Fraudulent impersonation deepfake

Using synthetic media to impersonate another person for financial or other unlawful gain.

Offence 3 — Election deepfake

Knowingly distributing materially deceptive synthetic media with intent to unlawfully influence an election.

Offence 4 — Deepfake evidence

Knowingly fabricating synthetic evidence and presenting it as authentic in judicial or official proceedings.

Offence 5 — Child sexual deepfake

Creation/distribution of prohibited synthetic sexual material depicting children.

Offence 6 — Malicious identity exploitation

Using an individual's likeness, voice or biometric characteristics for specified unlawful purposes.

24. Case Law on Personality Rights

Indian courts have increasingly dealt with unauthorised use of celebrity identity.

A significant example is:

Amitabh Bachchan v. Rajat Nagi & Ors.

The Delhi High Court granted protection against unauthorised commercial exploitation of Amitabh Bachchan's personality attributes.

The case illustrates the emerging Indian concept of personality rights, involving aspects such as:

• name;
• image;
• voice;
• likeness;
• persona.

This becomes especially significant with AI voice cloning and face generation.

25. Anil Kapoor v. Simply Life India

Anil Kapoor v. Simply Life India & Ors., Delhi High Court

The Delhi High Court granted broad protection concerning the unauthorised use of Anil Kapoor's personality attributes, including his name, image, likeness and related persona.

This is highly relevant to deepfake law because AI can reproduce:

• an actor's face;
• voice;
• mannerisms;
• signature expressions;
• digital persona.

The personality-rights approach provides a civil-law mechanism even where criminal law may not yet provide a perfectly tailored offence.

26. Recent Deepfake Litigation in India

Indian courts are increasingly confronting AI-generated impersonation and sexually explicit synthetic content.

For example, in 2026 the Delhi High Court granted interim relief to Tabu in proceedings concerning obscene AI-generated material using her identity and directed online platforms to remove targeted content.

The Delhi High Court has also recently granted relief to Janhvi Kapoor concerning allegedly fake/AI-generated obscene material and ordered removal from thousands of web pages.

These developments are important because they show that Indian courts are increasingly using personality rights, privacy, dignity and injunction/takedown mechanisms to respond to AI-generated impersonation even in the absence of a single comprehensive deepfake offence.

27. European Union — A Different Model

The EU has adopted a primarily transparency-based approach to many deepfakes rather than criminalising every deepfake.

Under Article 50 of the EU AI Act, providers and deployers have transparency obligations concerning AI-generated or manipulated content. AI-generated/manipulated content must be appropriately marked, and deepfakes are subject to specific disclosure requirements.

The EU's transparency obligations under Article 50 began applying on 2 August 2026.

This is an important comparative point:

The EU model does not simply say “deepfakes = crime.”

Instead, it combines:

transparency + detection + labelling + existing criminal/civil law.

28. United States — TAKE IT DOWN Act

The United States has moved toward specific federal criminal regulation of non-consensual intimate imagery, including AI-generated “digital forgeries.”

The TAKE IT DOWN Act defines a digital forgery to include an intimate visual depiction created through software, machine learning, AI or other technological means that is indistinguishable from an authentic depiction when viewed as a whole by a reasonable person.

This is significant because it specifically recognises:

AI-generated intimate imagery can itself be the unlawful material; there does not need to be an authentic underlying intimate photograph.

29. Comparative Legal Models

The U.S. federal approach has become particularly significant because Congress enacted the TAKE IT DOWN Act in 2025, addressing non-consensual intimate visual depictions including AI-generated digital forgeries.

30. Criminalisation vs Regulation

A sophisticated legal framework should use three layers.

Layer 1 — Criminal law

For serious intentional harm:

• sexual exploitation;
• fraud;
• extortion;
• child exploitation;
• electoral manipulation;
• fabricated evidence;
• identity-based criminal conduct.

Layer 2 — Civil remedies

For:

• personality-right violations;
• privacy;
• defamation;
• commercial exploitation;
• reputational harm.

Layer 3 — Platform regulation

For:

• labelling;
• provenance;
• detection;
• notice-and-action;
• preservation of evidence;
• removal of unlawful material.

This is much better than attempting to make criminal law solve every deepfake problem.

31. Safe-Harbour Problem

Platforms face a difficult question:

Should platforms be criminally liable merely because users upload deepfakes?

A strong intermediary framework should generally distinguish between:

passive hosting and knowing/active participation.

Otherwise platforms may respond by removing enormous amounts of lawful content simply to avoid liability.

That creates the risk of:

over-removal → censorship → chilling effect.

This concern fits directly with the constitutional principles recognised in Shreya Singhal.

32. Notice-and-Takedown for Deepfakes

For serious deepfakes, especially intimate imagery, a rapid takedown mechanism is highly desirable.

A victim should not have to:

identify thousands of URLs → file multiple lawsuits → wait months → prove the entire case before removal.

A modern law could provide:

1. victim complaint;
2. identity verification;
3. preliminary assessment;
4. rapid temporary removal;
5. preservation of evidence;
6. opportunity for uploader response;
7. final determination;
8. appeal mechanism.

This balances victim protection and due process.

33. Right to Explanation and Appeal

Deepfake regulation also creates the opposite risk.

Suppose a platform incorrectly labels a genuine political video as a deepfake.

That could affect:

• journalism;
• elections;
• political speech;
• public debate.

Therefore, regulation should include:

• notice;
• explanation;
• appeal;
• human review for serious decisions;
• restoration where content was wrongly removed.

34. Watermarking and Provenance

Criminalisation alone cannot solve the problem.

Technology can assist through:

Digital watermarking

AI-generated content contains a detectable signal.

Cryptographic provenance

The origin and editing history of media are recorded.

Content credentials

The system records:

created by X → edited by Y → published by Z.

Detection systems

AI tools attempt to identify synthetic media.

But detection is imperfect.

Therefore:

Legal liability should not depend solely on whether an AI detector says “deepfake.”

Courts should consider the totality of evidence.

35. Proposed Indian Deepfake Criminalisation Framework

A future Indian statute could contain the following structure.

Section 1 — Definition

Define synthetic/manipulated media.

Section 2 — Prohibited deepfake conduct

Creation, publication or distribution where specified unlawful intent exists.

Section 3 — Sexual deepfakes

Separate offence with enhanced punishment.

Section 4 — Child sexual deepfakes

Highest level of protection.

Section 5 — Fraudulent impersonation

AI-generated voice/video used to obtain money or property.

Section 6 — Election manipulation

Targeted offence for knowingly deceptive synthetic media intended to unlawfully interfere with electoral processes.

Section 7 — Judicial/evidentiary deepfakes

Fabricating or presenting synthetic evidence as genuine.

Section 8 — Extortion

Using a deepfake to threaten or demand money/benefit.

Section 9 — Platform obligations

Notice, action and preservation.

Section 10 — Labelling

Mandatory disclosure for certain synthetic content.

Section 11 — Exceptions

Explicitly protect:

• satire;
• parody;
• artistic works;
• research;
• journalism;
• education;
• legitimate political commentary;

subject to carefully defined conditions.

36. Exceptions Are Essential

A deepfake statute should expressly protect legitimate uses.

For example:

Political satire

A video showing a fictional politician saying something absurd may be protected parody.

Film

An actor digitally transformed into a historical figure should not be criminalised.

Education

A university may create an AI reconstruction for teaching.

Journalism

A broadcaster may use synthetic reconstruction to explain an event, provided it is clearly disclosed and not presented as genuine footage.

Research

Researchers may generate synthetic media to test detection systems.

The crucial distinction is:

Is the content being deceptively presented as authentic, and is there a legally recognised harmful purpose?

37. Mens Rea + Harm + Authenticity

A useful legal test would therefore be:

Step 1 — Synthetic?

Was the content generated or materially manipulated?

↓

Step 2 — Identifiable person?

Does it depict or impersonate an identifiable person?

↓

Step 3 — Material deception?

Would a reasonable person likely regard it as authentic?

↓

Step 4 — Knowledge/intent?

Did the creator/distributor know or intend that it would be deceptive?

↓

Step 5 — Unlawful purpose/harm?

Was it used for:

• sexual exploitation;
• fraud;
• extortion;
• unlawful electoral manipulation;
• defamation;
• harassment;
• fabricated evidence;
• another legally recognised harm?

↓

Step 6 — Protected expression?

Is it legitimate:

• satire;
• parody;
• art;
• journalism;
• education;
• research?

This produces a far more constitutionally defensible framework.

38. Important Case-Law Principles — Consolidated

39. Core Legal Principle

The most defensible legal approach is:

Do not criminalise falsity alone. Criminalise intentional, materially deceptive synthetic impersonation when it is connected to a clearly defined unlawful purpose or serious legally recognised harm.

This formulation protects both:

individual dignity and safety

and

constitutional freedom of expression.

40. Conclusion

Deepfake criminalisation is not simply a question of:

“Should AI-generated content be illegal?”

The real legal question is:

When does synthetic representation cross the constitutional and criminal-law boundary from protected expression into unlawful deception, exploitation or harm?

Indian law currently addresses different aspects through existing criminal law, the IT Act, privacy jurisprudence, intermediary regulation and personality rights rather than through one comprehensive Deepfake Prevention Act.

The constitutional cases of Puttaswamy, R. Rajagopal, Shreya Singhal and Anuradha Bhasin suggest that any future Indian deepfake statute should satisfy legality, precision, legitimate purpose, proportionality and procedural safeguards.

For particularly harmful categories—non-consensual intimate deepfakes, child sexual deepfakes, fraud, extortion, fabricated evidence and deliberate electoral deception—a stronger criminal response is justified.

The comparative trend is also clear: the EU is emphasising transparency and labelling, while the U.S. has moved toward specific federal criminalisation of non-consensual intimate digital forgeries.

Best formulation for an Indian legal research paper

“India should adopt a harm-based and intent-based deepfake criminalisation framework rather than a blanket prohibition, combining targeted criminal offences for serious harms with personality-rights remedies, rapid takedown mechanisms, AI-content labelling, provenance requirements, intermediary safeguards and explicit exemptions for legitimate satire, journalism, artistic expression, education and research.”