SEBI will shortly issue AI guidelines for India’s capital markets that require a kill switch, humans-in-the-loop controls and data controls, chairman Tuhin Kanta Pandey said in Mumbai on Wednesday. The framework will be tiered, with accountability and governance obligations scaled to how much risk a firm’s AI actually carries.

Speaking at the 23rd FICCI Capital Markets Conference 2026, Pandey framed the move as unavoidable rather than optional. “The question is not whether markets will use AI, the question is how we use it responsibly while preserving trust,” he said, according to news agency ANI.

Key takeaways

  • Three mandatory controls: a kill switch, humans-in-the-loop, and data controls.
  • Tiered, not one-size-fits-all: obligations scale with the risk of the deployment.
  • Not new policy from nowhere: SEBI floated a consultation paper on responsible AI/ML in the securities market in June 2025. These SEBI AI guidelines are that process landing.
  • Why now: 14.9 crore unique investors, market cap near 132% of GDP, and mutual fund assets around Rs 86 lakh crore.
  • Still unknown: no published date, no draft text, no penalty schedule.

What the SEBI AI guidelines will actually require

Pandey named three controls. Each is a specific engineering obligation, not a principle.

The three controls in SEBI’s AI guidelinesA kill switch to halt a model, humans-in-the-loop for consequential decisions, and data controls over training and inference data, all applied in proportion to risk.What SEBI says AI in the markets must haveThree named controls, applied on a tiered basis by riskKill switchA tested way to halt themodel immediately, withouttaking the business downwith it — plus the fallbackthat runs after you pull it.Humans in the loopA named, qualified personaccountable for what themodel does — with the skilland the authority tooverride it.Data controlsProvenance and protectionfor training and inferencedata — including what leavesyour perimeter when athird-party vendor is used.Applied on a TIERED basis — heavier obligations where the AI carries more riskSource: SEBI chairman Tuhin Kanta Pandey, FICCI Capital Markets Conference, 19 Aug 2026 (ANI) · Lapaas Voice

The wording matters. A kill switch is not a philosophical commitment to safety; it is a tested control that a firm must be able to demonstrate. So is a human in the loop. Regulators ask to see both.

Why SEBI is moving now

Pandey used the same speech to lay out how large the system being protected has become. Equity issuance crossed Rs 4.5 lakh crore in FY25-26, including about Rs 1.9 lakh crore raised through 366 IPOs. Corporate bond issuance exceeded Rs 9 lakh crore in FY26, with Rs 2.7 lakh crore already raised in the first four months of the current financial year.

Alternative Investment Fund investments stood at roughly Rs 7 lakh crore by the end of July 2026. Market capitalisation was around 132% of GDP. Mutual fund assets were about Rs 86 lakh crore, with SIPs accounting for more than a fifth of industry assets, across roughly 14.9 crore unique investors.

The scale SEBI’s AI guidelines have to coverMutual fund assets around 86 lakh crore rupees dwarf corporate bond issuance of 9 lakh crore, AIF investments of 7 lakh crore, equity issuance of 4.5 lakh crore and IPO proceeds of 1.9 lakh crore.The system SEBI is writing AI rules forRs lakh crore · figures cited by the SEBI chairman, 19 August 2026Mutual fund assets86.0Corporate bonds, FY269.0AIF investments7.0Equity issuance, FY25-264.5Of which 366 IPOs1.90Rs 43 lakh croreRs 86 lakh croreMarket cap ~132% of GDP · ~14.9 crore unique investors · Chart: Lapaas Voice

That is the argument for the SEBI AI guidelines in one paragraph. When a fifth of mutual fund assets arrive through automated monthly SIP instructions and nearly 15 crore people hold positions, a mis-specified model is no longer a firm’s private problem.

This has been coming since June 2025

The announcement is a landing, not a surprise. SEBI floated a consultation paper on guidelines for the responsible use of AI and ML in Indian securities markets in June 2025, and the obligations trailed then were more detailed than Wednesday’s three headline controls.

From consultation paper to guidelinesSEBI published a consultation paper on responsible AI and machine learning use in June 2025; in August 2026 the chairman said guidelines would be issued shortly.Roughly 14 months from consultation to rulesJune 2025Consultation paper onresponsible AI/ML useIndustry commentsScope and tiering debated19 Aug 2026Chairman: guidelinesto be issued “shortly”Sources: SEBI consultation paper (June 2025); ANI report of FICCI address · Timeline: Lapaas Voice

The consultation stage floated requirements that firms should expect to see carried through: skilled internal teams capable of real oversight rather than nominal sign-off, documented governance and fallback plans, enforceable agreements with third-party AI vendors, continuous monitoring, independent audits, and periodic reporting of model accuracy results to SEBI.

If those survive into the final text, the compliance burden is not the kill switch. It is the paperwork proving the kill switch works.

What a kill switch actually means for a broker

A kill switch obligation means a regulated firm must be able to stop a specific model from acting on the market within seconds, on command, without shutting down the platform around it — and must have a defined non-AI fallback that takes over the moment it is pulled. In practice that forces AI to be built as a separable component with its own isolation boundary, rather than woven invisibly through order routing, surveillance and client servicing.

That is an architecture requirement disguised as a safety requirement, and it is the expensive part. A firm that has embedded model calls deep inside its execution path cannot isolate them later without rework.

The human-in-the-loop requirement carries a similar sting. Pandey stressed accountability and governance, and the consultation stage pointed to firms needing genuinely skilled internal teams. A compliance officer who cannot interrogate a model is not oversight; they are a signature. Firms that outsourced their AI wholesale will find that harder to satisfy than firms that built in-house.

Who this lands on

  • Brokers running algo execution — the clearest target, and the ones with the most to re-architect.
  • Asset managers using models for portfolio construction, risk or research.
  • Registered investment advisers and research analysts generating recommendations with AI assistance.
  • Fintech distribution platforms using AI for onboarding, suitability or nudges — where a biased model touches retail investors directly.
  • Third-party AI vendors, indirectly, through the agreements their regulated clients now need.

Pandey was explicit that AI cuts both ways for a regulator. It can strengthen market surveillance, risk assessment, fraud detection and investor servicing. It also creates risks around opacity, bias, cybersecurity, data protection and accountability. The guidelines are an attempt to keep the first list while bounding the second.

India’s financial regulators have been moving in step on this kind of operational tightening — see the RBI’s plan to reset floating rate loans quarterly from 2027 and its tightening around gold lending by NBFCs.

What SEBI has not said

Four things are still missing, and they are the ones that determine cost.

  • A date. “Shortly” is not a deadline.
  • The tiering test. What puts a deployment in the high-risk tier — order value, client count, autonomy, asset class? This single definition decides who pays for full compliance.
  • Transition time. Whether existing systems get a grandfathering window.
  • Consequences. No penalty schedule has been indicated.

Until the text is published, the honest read is that direction is confirmed and detail is not.

Frequently asked questions

What are the SEBI AI guidelines?

They are a forthcoming framework for the responsible use of artificial intelligence and machine learning by participants in India’s securities markets. SEBI chairman Tuhin Kanta Pandey said on 19 August 2026 that they would require kill-switch mechanisms, humans-in-the-loop controls and data controls, applied on a tiered basis.

When will the SEBI AI guidelines be issued?

No date has been announced. Pandey said only that SEBI would be issuing them “shortly”. The underlying consultation paper dates to June 2025.

Who has to comply?

Market participants using AI or ML — in practice brokers, asset managers, registered advisers, research analysts and platforms, with obligations expected to scale by risk tier rather than applying uniformly.

Does this ban algorithmic trading?

No. Nothing announced restricts the use of AI or algorithms. The framework governs how they are controlled, supervised and documented.

The bottom line

SEBI is not trying to slow AI down in Indian markets. It is trying to make sure that when a model misbehaves, somebody can reach the switch and somebody’s name is on the decision.

For firms that treated AI as an experiment bolted onto production, the SEBI AI guidelines will be an expensive re-architecture. For firms that built it as a governed component from the start, this is mostly documentation. The gap between those two groups is about to become visible.

Reported remarks sourced from news agency ANI and BusinessLine. Lapaas Voice will update this article when SEBI publishes the guidelines.

Get the day’s top stories in your inbox

One concise email. No spam, unsubscribe anytime.