This is a MapMyTimes opinion piece. It responds to remarks by Securities and Exchange Board of India (SEBI) Whole Time Member KVR Murty at a World Federation of Exchanges event in Mumbai on 19 November 2026, where he said SEBI is preparing guidelines on the responsible use of artificial intelligence (AI) in the capital markets. MapMyTimes reported the speech here.
Regulators rarely admit in public that they do not have the answers. Murty did. In his published address, he described an AI agent as something given "a goal, not a fixed set of rules", one that "can change how it trades as it learns", and asked: "how do you approve a strategy that rewrites itself?" He then added: "We don't have every answer yet, and I doubt any regulator in this room does."
That honesty is useful. It also sets the bar for the guidelines SEBI now says it is preparing. They will be judged on whether they turn one sentence from the same speech into rules that work: "every agent, however intelligent, must be traceable, and someone must answer for what it does."
Why does the question matter now?
It matters because the tools have already reached ordinary investors. Murty said AI "now helps ordinary investors build their own algos". India's current answer to retail algorithmic trading, as he described it, rests on three checks: every strategy is approved by an exchange, every strategy carries an identifier, and a broker stands behind every order. Those checks assume a strategy stays the same after it is approved. A self-learning agent breaks that assumption.
The second risk is collective. Murty asked what happens "when a million investors run agents built on the same few models and the same data, and all head for the exit at once". A market where many participants rely on the same handful of models could see their decisions line up in a stressful moment. He also called third-party dependencies, including cloud, data and model providers, "a systemic channel of risk".
What should the guidelines do?
Three things would give "traceable" and "accountable" real meaning.
- Name the answerable party. When an AI agent places orders or gives advice, investors should be able to tell which regulated entity is responsible for it. SEBI's own June 2025 consultation paper, as reported by Business Today, proposed that senior management be accountable for AI oversight. The final rules should keep that clarity rather than dilute it.
- Tell investors when AI is involved. The same 2025 draft proposed plain-language disclosure of a model's purpose, limits and fees where AI directly affects clients, and logs of inputs and outputs kept for at least five years. Disclosure and record-keeping are what make tracing possible after something goes wrong.
- Watch for crowding. If concentration on a few models is a market-wide risk, as Murty suggested, the rules should ask firms to report which outside models and providers their client-facing tools rely on, so the regulator can see when too many investors are leaning on the same system.
Is there a case for going slower?
Yes, and it deserves a fair hearing. Rules written for a technology that is still changing can age badly, and Murty himself said no regulator has every answer. Heavy compliance costs could push smaller fintech firms out, or nudge retail investors towards tools run from outside India's regulatory reach. India also already has a working framework for retail algorithmic trading, and some will argue that SEBI should extend it before writing something new. Murty's own test for innovation, which asks whether it is useful, whether it reduces risk or cost, and whether it can scale, could be read as a reason for patience.
These concerns argue for careful design, not for delay. Murty described AI that advises retail investors or trades on its own as "still experimental", which is exactly where clear lines of responsibility matter most. Principles that stay fixed while technical details are reviewed over time, a lighter path for internal uses such as surveillance (which the 2025 draft also proposed), and a public consultation on the actual text would address most of the worry about stifling innovation.
What would a good outcome look like?
SEBI has not given a timeline for the guidelines. When the text arrives, the simplest test will be this: if an AI agent loses an investor money or helps move a market, can that investor and the regulator find out what it did and who is responsible? Murty ended his speech by saying technology "can strengthen" trust in markets but "cannot substitute for it". Guidelines that make every agent traceable to someone accountable would put that idea into practice.




