The Indian government today proposed strict new rules that would require the clear and conspicuous labelling of all content generated by AI. Citing “risks growing” of deceptive deepfakes, the Ministry of Electronics and Information Technology (MeitY) circulated draft modifications to the 2021 IT Rules, hoping to hold social media firms and AI producers responsible.
The move comes as a direct response to the widespread use of AI-driven misinformation, particularly during the 2024 general elections. While a 2023 deepfake video of an actress drew public attention, there were massive political deepfake videos released this year, including AI-generated videos of actors Aamir Khan and Ranveer Singh appearing to endorse political parties and a video of politician Asaduddin Owaisi fabricated to show him singing Hindu devotional songs, which caused the regulating bodies to come up front.
Major points of the draft regulations
The new proposed amendments bring forth the new, concise definition of “synthetically generated information” and set forth the three-tiered system of responsibility:
Mandatory User Declaration: Significant social media intermediaries (platforms with over 5 million users) must require users to declare whether any content they upload is AI-generated.
Validation of Platforms: Platforms will not be able to solely depend upon the users. They must adopt reasonable and proportionate technical measures and individually authenticate these claims.
Prominent Labelling: All content authored by AI should be clearly and prominently labelled. The draft specifies that this label should cover the surface area of not less than 10 per cent for visual content and be included in the initial 10 per cent of the duration for audio content.
The broader AI strategy
As Part of the Broader AI Plan, “Safe and Trusted AI” is an agreed key pillar of the government's multi-billion-dollar “India AI Mission” aimed at stimulating innovation. In late 2024, the government issued a call for proposals for the creation of “watermarking & labelling” and “deepfake detection tools.”
