India’s AI Rules Are Arriving in Silos: One Company, Multiple Rulebooks
While India does not yet have an AI Act, it is quietly developing another equally challenging one: a separate rulebook for each regulator. ASCI, SEBI and the I&B Ministry have each issued or announced rules in their respective areas since the end of September. The tests don’t match, and a business whose AI interacts with markets, advertising and news content now has to meet all three.
Markets: kill switches
On 10 October, SEBI Chairman Tuhin Kanta Pandey stated that SEBI’s AI/ML guidelines will be tiered and will include accountability, data controls, kill switches and human oversight, in line with the supervisory toolkit of IOSCO. He had previously flagged them on 19 August. His message was clear: technology cannot dilute accountability, and the regulated is responsible for the outcome of AI, data quality and model governance.
Banking: model risk
RBI’s June draft guidance on model risk management includes all models used by the regulated entity, including AI/ML and third-party models. It suggests that the decisions made by AI systems should be subject to board-approved governance and human oversight. Comments were closed on 24 July, and the final text is pending. A group with both a SEBI-registered and an RBI-regulated arm could find two identical tests for the same model.
Broadcasting: AI stays assistive
The I&B Ministry on 21 October advised the private channels to use AI, including generative AI, as an assisting tool, not to affect the accuracy, authenticity or sensitivity of news, and to adhere to the Programme Code under the Cable Television Networks (Regulation) Act, 1995. Any violation of the code can lead to action under the existing framework, and AI-generated graphics are no exception to this rule, the ministry warned.
Two labelling tests: Advertising and platforms
The guidelines for synthetic ads, issued by ASCI on 29 September, will come into effect on 1 January 2027. As Khaitan & Co notes, ASCI requires labels only where content ‘materially influences’ consumer decisions, whereas the IT Rules amendment in force since 20 February requires intermediaries to label synthetically generated audio-visual content, backed by three-hour takedown timelines.
One Creative, two tests: An AI-generated “customer” praising a product may need an ASCI label because it influences buying decisions, while an AI-edited background may not. However, platforms may still need to label both. Thus, the same advertisement can face two different tests.
Safety testing and the law ahead
On 15 October, Anthropic’s India policy head said the company is in talks with MeitY to make models available to the IndiaAI Safety Institute, and wants binding transparency rules. The same report notes the Centre is preparing a dedicated AI law, a turn from the November 2025 AI Governance Guidelines, which saw no immediate need for such a standalone law. Minister Ashwini Vaishnaw on 8 October stated that a consultation paper on AI safety and deepfakes would follow within a month, adding that AI requires a techno-legal solution with industry on board, not legislation alone.
What this means
While the formats of the rulebooks vary, they share three concepts: humans remain in the loop, accountability remains with the deployer, and AI-generated content must be identifiable. Companies should have one AI inventory per regulator’s test and ask a simple question of each system: who can switch it off and how fast? And the consultation paper is the moment to ask for harmonised definitions across regulators.
