← Insights and Updates Media and Technology

Personality rights and the limits of the AI safe harbour

As synthetic likenesses proliferate, the intermediary defence is being tested. Where Indian law currently stands on impersonation, consent and platform liability.

The problem

Generative tools now make it trivial to produce a convincing video or audio recording of a person saying or doing something they never said or did. The applications range from the merely commercial, an unauthorised endorsement, to the seriously harmful: fabricated confessions, synthetic intimate imagery, and manufactured evidence.

The rights engaged

Indian law protects personality through several overlapping routes rather than a single statute. Publicity rights, recognised in a line of decisions, protect the commercial value in a person's name, image, likeness and voice. Passing off protects against misrepresentation of endorsement. Defamation protects reputation. Article 21 protects privacy, and following the recognition of informational privacy as a fundamental right, the unauthorised use of a person's likeness engages constitutional protection. The Digital Personal Data Protection Act, 2023 adds a consent architecture around personal data, and a photograph or voice recording of an identifiable person is personal data.

The safe harbour

Section 79 of the Information Technology Act, 2000 exempts an intermediary from liability for third-party content, provided its function is limited to providing access, it does not initiate the transmission or select or modify the content, and it observes due diligence including the requirements of the Intermediary Guidelines. The exemption is lost where the intermediary has conspired, abetted or induced the unlawful act, or where, upon receiving actual knowledge, it fails to expeditiously remove the material.

Where the tension lies

The safe harbour was designed for a platform that hosts what its users upload. It sits awkwardly on a platform that generates the content itself in response to a prompt. A model that synthesises a likeness is not, on any natural reading, merely providing access to information made available by a third party; it has selected and modified, and arguably initiated. Whether generative services can claim Section 79 protection at all is, at present, an open question, and one on which litigants should expect the argument rather than assume the answer.

Practical steps

For individuals whose likeness has been misused: preserve the material with timestamps and URLs before it is taken down, issue a takedown notice under the Intermediary Guidelines, and consider a John Doe injunction where the uploader is unidentified, which is the usual case. Speed matters more than perfection. For platforms and businesses deploying generative tools: obtain express, recorded, purpose-specific consent before synthesising any identifiable person, and do not rely on a general terms-of-service clause to supply it.

Request a Consultation