SEBI Again Says AI Rules for Markets Are Coming 'Shortly' -- Now With Kill Switches, Human Oversight and an IOSCO Toolkit to Align With

Speaking at the second Capital Market Confluence 2026, organised by the BSE Brokers' Forum, on Saturday, October 10, 2026, Pandey said the AI framework "would follow a tiered approach with clear accountability, data controls, kill switches and human oversight," ANI reported. It will "also align with the International Organization of Securities Commissions' supervisory toolkit for AI to support an agile governance framework." His line on liability was blunt: "Technology cannot dilute accountability. Regulated entities remain responsible for AI outcomes, data quality, model governance and cyber resilience."

Not the first time

The promise itself is not new. At the FICCI Capital Markets Conference on August 19, Pandey said: "We will shortly be issuing guidelines for responsible use of AI/ML in our markets," with "kill-switch and humans-in-the-loop controls along with data controls." Moneylife's report of that speech added that every regulated entity would stay accountable for AI tools "irrespective of whether the technology is developed internally or procured from an external provider." The underlying consultation paper, "guidelines for responsible usage of AI/ML in Indian Securities Markets," was published on June 20, 2025. Nearly sixteen months later, no final circular has been issued. What October adds is the explicit IOSCO alignment, which suggests SEBI wants rules that look familiar to international supervisors and to foreign firms operating in India.

What the rules are likely to cover

The 2025 paper, as summarised by Taxmann, proposed principles of equality, accountability, transparency and safety, and disclosure of AI use in algorithmic trading, asset management, portfolio management and advisory and research services. It also proposed records of model development, testing, deployment and monitoring. In August, Pandey listed both the uses – surveillance, risk assessment, fraud detection and investor services – and the risks: "opacity, bias, cybersecurity, data protection and accountability."

Why it matters for firms building with AI

• Vendor models are no shield. If the August framing holds, a broker using a third-party model for KYC, client chat or trade surveillance carries the same accountability as one that built its own.
• Kill switches imply architecture. A requirement to halt an AI system means firms need a way to switch off the model path without stopping the business – a fallback to rules or people, tested in advance.
• Tiering decides cost. A tiered approach means obligations scale with risk; which uses land in the top tier (algorithmic trading and advice are the obvious candidates) will decide most of the compliance burden.

The scale is large. Pandey put market capitalisation at around Rs 472 lakh crore, with more than 15 crore unique investors and mutual fund assets of around Rs 87 lakh crore.

What remains uncertain. The timing, given two "shortly" statements two months apart; how tiers will be defined; whether disclosure goes to clients as well as to SEBI; and how the rules will treat AI agents that take actions, not just models that score or advise.

Analysis: Indian market firms should not wait for the circular. The direction has been stable across three statements: firm-level accountability regardless of vendor, human oversight, data controls and a working off switch. An inventory of where AI touches clients and orders, plus a tested manual fallback for each, will be needed under any version of the final text.