Key Details
In an address at FICCI’s Annual Capital Markets Conference, SEBI Chairman Tuhin Kanta Pandey distinguished measures already introduced from reforms still under consultation or review. The clearest new regulatory signal is a forthcoming tiered framework for responsible use of AI and machine learning.
Status | Measures Highlighted |
|---|---|
Forthcoming | AI/ML guidelines requiring clear accountability, data controls, human oversight and a “kill switch” |
Under consultation or consideration | New portfolio-management products, wider commodity-derivatives access for FPIs and Depository Receipts for REIT and InvIT units |
Under review | Securities lending and short selling, settlement regulations, SME IPOs, LODR, delisting and accredited-investor rules |
Already implemented | Faster AIF launches through GARUDA, simplified access through SWAGAT-FI and risk-based inspection of intermediaries |
AI Responsibility Cannot Be Outsourced to a Vendor
SEBI-regulated entities will remain responsible for every AI or machine-learning system they use, whether built internally or purchased from a third party. That responsibility covers the privacy, security and integrity of investor data, as well as the system’s outputs.
This closes an important accountability gap. A broker, mutual fund, exchange or other intermediary would not be able to attribute a harmful recommendation, discriminatory outcome or data breach solely to the technology provider.
The proposed tiered approach indicates that regulatory controls may vary according to risk. An AI tool used for routine customer support, for example, does not raise the same concerns as one influencing investment decisions, trading activity, fraud detection or access to financial products. The detailed categories and obligations will become clear only when SEBI issues the guidelines.
“Kill Switch” and Human Oversight Move the Debate towards Operational Controls
The speech identifies two concrete safeguards:
a kill switch, allowing an AI system to be stopped when it behaves unexpectedly or creates unacceptable risk; and
humans in the loop, requiring meaningful human oversight over specified processes or decisions.
These safeguards move beyond general principles of responsible AI. Their effectiveness will depend on practical questions: who is authorised to intervene, which events trigger a shutdown and how quickly a human must review a decision, as well as whether firms must retain records explaining model outputs and overrides.
SEBI is already using AI through Project SUDARSAN and R(AI)DAR to detect suspicious financial promotions and potentially misleading advertisements. The regulator is therefore developing rules for technology that it is also applying within market oversight.
Capital-Market Reforms Target New Investors and Financing Channels
The wider policy agenda seeks to expand both participation and the range of assets available through regulated markets.
Among the proposals discussed are:
allowing portfolio managers to invest client funds in foreign securities;
creating a mutual-fund-only portfolio-management service for mass-affluent investors;
simplifying digital onboarding for persons resident outside India;
widening FPI participation in non-agricultural commodity derivatives;
exploring tokenisation of corporate bonds;
developing a Credit Risk-o-Meter for retail debt investors; and
allowing greater REIT and InvIT exposure to projects under construction within prudential limits.
These measures are at different stages. Some are consultation proposals, while others are being considered or reviewed. The speech does not make them final regulatory requirements.
SEBI Is Moving Supervision towards Risk rather than Routine
The regulator has reduced repetitive inspections of compliant entities and is directing greater supervisory attention towards intermediaries with higher risk scores, alerts or adverse market intelligence.
The same logic appears in other reforms: lighter processes for sophisticated investors, faster AIF scheme launches and simplified access for identified low-risk foreign investors. This approach can reduce unnecessary compliance costs, but it places greater importance on the quality of SEBI’s risk indicators and early detection of misconduct.
Markets Have Scale; SEBI Is Now Seeking Greater Depth
The speech places the reform agenda against a considerably larger domestic market:
equity issuance exceeded ₹4.5 trillion in FY2025–26, including around ₹1.9 trillion raised through 366 IPOs;
corporate bond issuance crossed ₹9 trillion during the year;
mutual fund assets reached approximately ₹86 trillion;
India had around 149 million unique investors; and
market capitalisation stood at roughly 132% of GDP.
SEBI’s next-stage agenda is therefore not simply to increase transaction volumes. It seeks to widen participation, deepen debt and pooled-investment markets, finance newer sectors and use technology without weakening trust.
What Is an AI Kill Switch?
An AI kill switch is a control that allows an organisation to suspend or disable an AI system when it produces unsafe, unlawful or unreliable results. In financial markets, it could be used to stop automated activity before faulty outputs spread across investor accounts, transactions or critical systems. Its value depends on clear triggers, authorised decision-makers and tested fallback procedures.
Policy Relevance
The proposed AI guidelines could establish one of India’s more operational sector-specific AI governance frameworks. Unlike broad statements of principle, they are expected to connect responsibility with identifiable controls inside regulated financial institutions.
Three design choices will shape their effect:
Risk classification: obligations need to reflect whether AI is performing an administrative task, influencing a customer, assessing eligibility or interacting with market activity.
Third-party accountability: regulated entities will require audit rights, incident information and data protections in contracts with technology vendors.
Human intervention: oversight must be timely and meaningful; nominal approval after an automated decision has already caused harm would provide little protection.
The framework’s credibility will ultimately depend on whether SEBI specifies testing, record-keeping, incident reporting and supervisory expectations alongside the announced controls.
Relevant Question for Policy Stakeholders: Which uses of AI in securities markets should require prior human approval or immediate shutdown capability, particularly when the underlying model is controlled by an external vendor?
Follow the Full Speech Here: Address by Shri Tuhin Kanta Pandey, Chairman, SEBI, at FICCI’s 23rd Annual Capital Markets Conference