AI Consultation Paper: Experts Want Rules On Agents, Deepfakes, Schools

The Narendra Modi government will issue a consultation paper on artificial intelligence (AI) in the coming weeks, IT Minister Ashwini Vaishnaw said at the NDTV World Summit 2026. "That consultation paper will be where we will seek all the inputs from society, from citizens, from industries, and stakeholders," he said.Vaishnaw stressed that some principles are already settled - AI must be human-centric, and control over agents, models, outcomes and processes must remain in human hands. "There has to be some equivalent of stopping things if they go out of control," he said.READ | AI Law In India Soon? What Ashwini Vaishnaw Said At NDTV World SummitWe asked experts how they would like to see AI regulations in India take shape. Make "Human Control" RealDr Srinivas Padmanabhuni, a PhD in Artificial Intelligence (AI) from University of Alberta, Edmonton, Canada and the CTO of AiEnsured, told NDTV that human control has to be explicitly spelled out in technical terms. "Human control requires blocking new tool calls, revoking access, cancelling pending jobs and displaying action statuses," he said.Padmanabhuni added that stopping an agent is not the same as undoing what it has done, so agents also need duplicate protection, audit trails and recovery procedures. He also highlighted that these controls must be tested mid-task and across handoffs.The AI consultation paper must define what an AI may do without approval and who is accountable, Padmanabhuni said. Requirements should "scale by risk, not the 'agent' label", with autonomy defined by observable powers such as the tools used, decisions executed and permission limits. On where it's most urgent, Padmanabhuni said banking and financial services need rules first, given the combination of money, sensitive data and scale. He wants explicit transaction authority, strict operational limits, traceable decisions, human escalation paths and clear customer appeal processes.Start With The Harms Already HereAI advisor Sagar Vishnoi, co-founder of Future Shift Labs, told NDTV that India is "already two to three years late" and the AI consultation paper should begin with harms already happening, take it from there and look ahead.He listed four areas:Deepfakes and synthetic media, which are fuelling fraud, harassment and election manipulation.Agentic AI, where agents that transact, send emails and access data on someone's behalf create an accountability gap. The paper should say who is answerable: the developer, the deployer, the user or the company.Children, including the mental health effects of companion bots and addictive design.Exclusion, where AI used in welfare, healthcare, credit and hiring could shut out people based on religion, caste, tribe or gender.Vishnoi also called for an incident-reporting mechanism so India's AI ecosystem can learn from evidence.Don't Forget The Small BusinessMany Indian firms deploying AI agents have four or five people, or one person handling everything, Vishnoi noted. He urges for a free playbook and ready-to-use checklist in Indian languages, covering disclosure to customers, human review, data handling and a kill switch, written in plain, simple language.While low-risk uses like drafting and scheduling need basic transparency, agents that handle data or money should be treated as high-risk, he said. On liability, Vishnoi said responsibility should be shared between developer and small business when licensed models are used, and the whole risk should not fall on the business when the developer controls the system.Fix How We TeachAI educator Ansh Mehra, founder of The Cutting Edge Group, said the issue needs to start getting addressed in classrooms. "Schools and universities build curricula with great titles, but the underlying syllabus is not as nuanced as it needs to be," he said, adding that marks cannot measure the most valuable subjects.If Mehra were designing a school curriculum, it would begin with reading, articulation, metacognition, pattern finding, observation and the ability to find overlaps between subjects. These skills will help build "real intelligence" as artificial intelligence keeps improving.The consultation paper is expected in the coming weeks. The experts' wishlist suggests the debate will turn on whether "human control" is defined in enough detail to be enforced, and on who is held responsible when an agent goes wrong or rogue.