SEBI is the last major Indian financial regulator without a comprehensive AI/ML governance circular. That gap is about to close. SEBI Chairman Tuhin Kanta Pandey announced at FICCI's 23rd Annual Capital Markets Conference on August 19, 2026 that formal AI/ML guidelines are forthcoming — with mandatory kill switches, human oversight controls, tiered risk-based rules, and full third-party vendor accountability.
The announcement confirms the direction of travel. But it is not the circular. The formal binding guidelines following the June 2025 consultation paper were not published as of October 2026. For SEBI-regulated entities — exchanges, depositories, brokers, mutual funds, investment advisors, portfolio managers, merchant bankers, credit rating agencies — the question is whether to wait for the circular or build governance infrastructure now.
The answer is clear: three binding instruments are already in effect (Regulation 16C, the algo trading circular, the May 2026 cybersecurity circular), and the forthcoming guidelines will build on these foundations. Entities building compliance infrastructure now will require minor adaptation when the final circular issues. Entities starting from scratch after the circular will face a compressed timeline.
What Is Already Binding vs. What Is Still Forthcoming
Already binding: Regulation 16C (sole accountability for AI/ML outputs and data — ALL regulated entities, February 2025); Algo trading Algo-ID, white box/black box classification, five-year audit trail (mandatory April 1, 2026); SEBI Cybersecurity Circular SEBI/HO/13/19/12(1)2026-ITD-1_CIMGI/10873/2026 AI defense mandates (May 2026).
Still forthcoming (announced, not yet published): Comprehensive AI/ML governance circular with kill switch requirements, HITL human oversight specifications, tiered risk framework (criteria for each tier), board-level AI governance framework, model audit requirements, explainability standards, data governance requirements beyond Regulation 16C's accountability baseline. Timeline: SEBI Chairman indicated "soon" at August 2026; likely Q4 2026 or Q1 2027 publication.
Four-Year Regulatory Stack: From 2019 Disclosure to 2026 Governance
Mutual Fund AI/ML Quarterly Disclosure Requirement
SEBI's first AI-specific rule: mandatory quarterly reporting of AI/ML applications by mutual funds. Submitted to AMFI within 15 days of quarter-end; AMFI consolidated and sent to SEBI within 30 days. Inventory exercise only — no governance mandate, no accountability rule, no technical requirements. Significant as the first acknowledgment that AI in Indian capital markets required regulatory attention.
Scope: Mutual funds onlyRegulation 16C — Binding Sole Accountability for AI/ML
SEBI (Intermediaries) (Amendment) Regulations, 2025 inserted Regulation 16C into the SEBI (Intermediaries) Regulations, 2008 with parallel amendments to Stock Exchanges and Clearing Corporations Regulations. Notified February 6, 2025; published February 14, 2025. Immediate legal force. Scope: all SEBI-regulated entities using AI/ML tools, in-house or third-party. Three obligations: (1) solely responsible for privacy, security and integrity of investor/stakeholder data; (2) solely responsible for all outputs arising from AI/ML tools; (3) solely responsible for compliance with all applicable laws. Third-party AI services are deemed to be provided by the intermediary, not the vendor. Harmful outputs cannot be attributed to the technology provider.
Scope: ALL regulated entities · Currently bindingAlgo Trading Circular — Algo-ID, White Box/Black Box, Five-Year Audit Trail
Circular SEBI/HO/MIRSD/MIRSD-PoD/P/CIR/2025/0000013 (February 4, 2025). Industry Standards Forum guidelines finalized April 1, 2025; effective August 1, 2025; mandatory compliance April 1, 2026. This is the most specific AI governance circular SEBI has issued: every algo order must carry a unique exchange-assigned Algo-ID; all AI/ML-using algorithmic strategies must be classified as white box (transparent, interpretable logic) or black box (non-transparent logic); black-box providers must register as SEBI Research Analysts; five-year audit trail mandatory for all automated trades; static IP, two-factor authentication, OAuth mandatory for algo infrastructure; exchange pre-approval required before any AI/ML algo deployment; retail investors with self-developed algos exceeding order-per-second thresholds must register with exchanges through their broker.
SEBI/HO/MIRSD/MIRSD-PoD/P/CIR/2025/0000013 · Mandatory from April 1, 2026Consultation Paper: Guidelines for Responsible Usage of AI/ML in Indian Securities Markets
Published June 20, 2025. Comment deadline July 11, 2025. Scope: Market Infrastructure Institutions, market intermediaries, mutual funds. Proposed a principles-based framework with five core pillars (see below). Six specific measures: senior management accountability, robust testing and monitoring, skilled oversight teams, third-party provider management, transparency in AI/ML use, high-quality unbiased data. Board-approved AI Governance Framework required. Mandatory periodic audits and model validations for bias detection. AI/ML model explainability requirements. Industry feedback: too vague on human oversight model (HITL vs. HOTL vs. HIC not defined), one-size-fits-all approach, no technical thresholds, unclear liability among developers/deployers/boards.
Consultation paper — NOT yet final circularCybersecurity Circular — AI Defense Mandates for All 10,000+ Regulated Entities
Circular SEBI/HO/13/19/12(1)2026-ITD-1_CIMGI/10873/2026 (May 5, 2026). Scope: all SEBI-regulated entities (exchanges, depositories, brokers, mutual funds, custodians, credit rating agencies, merchant bankers, portfolio managers, investment advisors — 10,000+ entities). Focus: AI-driven cybersecurity risks, specifically advanced AI vulnerability detection tools. Key AI mandates in Annexure-A: AI-based vulnerability assessment tools required; scenario-based AI risk testing; long-term plans for AI in autonomous/agentic threat mitigation; AI-augmented Security Operations Centre transformation; Zero Trust architecture. IT committees must ratify long-term AI strategy. This circular mandates AI adoption for defense while explicitly addressing AI-enabled attacks as the primary threat — SEBI acknowledging that AI is simultaneously the solution and the risk.
SEBI/HO/13/19/12(1)2026-ITD-1_CIMGI/10873/2026 · Currently bindingChairman Pandey at FICCI — Kill Switch, Human Oversight, Tiered Rules Announced
FICCI 23rd Annual Capital Markets Conference, Mumbai, August 19, 2026. Theme: "Capital Markets as the Gateway to India's Future: AI, Financialization and New Investment Frontiers." Pandey announced four forthcoming requirements: (1) mandatory kill switch for unexpected AI/ML behavior; (2) humans-in-the-loop controls for specified processes and financial decisions; (3) tiered risk-based rules (regulatory-lite for internal AI, stricter for customer-facing market-impact AI); (4) full third-party vendor accountability extending Regulation 16C. SEBI's own AI tools highlighted: SUDARSAN (suspicious financial promotion detection), R(AI)DAR (misleading advertisement detection), Cyber Suraksha Portal. Integration with IOSCO AI supervisory toolkit confirmed. Quote: "The question is not whether markets will use AI, the question is how we use it responsibly while preserving trust."
Speech only — formal circular forthcoming, not yet published as of October 2026Five Pillars of SEBI's June 2025 Consultation Paper
Model Governance
Board-approved AI governance framework. Senior management accountability. Robust model testing and monitoring before and after deployment. Periodic model validations, bias detection audits. Skilled oversight teams with AI domain expertise.
Investor Protection and Disclosure
Transparency requirements for AI/ML use with investors and clients. Explainability of AI decisions affecting investors. Adverse decision explanations. Mandatory disclosure when AI is used in investment recommendations, fraud detection affecting investor accounts, or portfolio management.
Testing Frameworks
Pre-deployment testing for accuracy, bias, robustness, and adversarial inputs. Ongoing monitoring post-deployment. Validation against real-world performance. Model drift detection. Stress testing under adverse market conditions. Backtesting for algo trading AI.
Fairness and Bias
Bias detection and mitigation in AI systems affecting investor access, credit, or product eligibility. DPDP Act alignment for data used in AI. India-specific fairness testing considerations: investor socioeconomic background, geographic region, investment experience level. Periodic audits for discriminatory outcomes.
Data Privacy and Cybersecurity
Data governance policies for AI training and inference data. DPDP Act 2023 alignment for investor personal data in AI systems. Cybersecurity requirements for AI infrastructure. Integration with May 2026 cybersecurity circular AI mandates. Third-party data provider due diligence.
August 2026 Announcement: Four Key Requirements Decoded
Mandatory Kill Switch
Regulated entities must maintain an immediate suspension capability for AI/ML systems that behave unexpectedly, produce unsafe results, or breach defined risk thresholds. Prevents faulty automated actions from propagating across investor accounts or market infrastructure. The formal circular will specify: authorization protocols (who can activate), shutdown triggers (behavioral criteria requiring activation), and response timelines. Algo trading already has trade halt mechanisms; the kill switch requirement extends this concept to all AI/ML deployments including fraud detection AI, customer service AI, and investment recommendation AI. The CERT-In Blueprint's kill switch concept (for banking AI) and SEBI's forthcoming requirement reflect a converging expectation across India's financial regulators that high-impact AI must have a manual override capability.
Human Oversight (HITL)
Mandatory 'humans-in-the-loop' for specified processes and financial decisions. Three oversight models exist in practice: HITL (human approves each AI decision before it executes), HOTL (human monitors and can intervene; AI executes unless overridden), HIC (human sets policy; AI executes within defined boundaries). The formal circular will define which model applies where. August 2026 speech critics noted this distinction was also absent from the June 2025 consultation paper — the final circular must resolve it. For high-impact decisions (a fraud suspension of an investor account, a large automated trade, a margin call AI decision), HITL is the highest standard. For surveillance and monitoring AI, HOTL may be appropriate. The final tiered rules will likely map oversight model to risk tier.
Tiered Risk-Based Rules
Regulatory-lite framework for AI that assists with internal compliance, surveillance operations, and cybersecurity (lower investor impact, lower market impact). Stricter requirements for AI that influences investment decisions, trading activity, fraud detection affecting accounts, or financial product access (direct investor impact, potential market impact). Tiering methodology — by firm size versus by AI risk level — was not defined in the August speech. The June 2025 consultation paper received industry feedback that a one-size-fits-all approach was unworkable for SEBI-regulated entities that range from large stock exchanges to small investment advisors. The final tiered framework will resolve this, likely drawing from the proportionality approach in RBI FREE-AI (which scales obligations to the bank's AI portfolio size and risk) and IRDAI's working group proportionality mandate (insurance-equivalent of scaling by insurer size).
Third-Party Vendor Accountability
Extends Regulation 16C's sole accountability to operational controls: regulated entities must secure audit rights, incident information rights, and data protection obligations through vendor contracts. Harmful outcomes from third-party AI cannot be attributed solely to the technology provider — the regulated entity is liable for any violation of securities laws caused by vendor AI it deployed. Practical implications: vendor contracts must include: right to audit AI models and their outputs; incident notification obligations when the vendor's AI malfunctions; explicit data protection and DPDP Act compliance warranties; performance SLAs with remediation rights when AI fails governance tests. Entities that accepted vendor AI contracts without these provisions before Regulation 16C need to renegotiate, or they retain liability without operational recourse.
What SEBI-Regulated Entities Should Do Now
| Action Item | Regulatory Basis | Timeline | Status |
|---|---|---|---|
| AI/ML inventory under Regulation 16C — document every AI/ML tool in use, whether in-house or third-party, with named accountable officer per tool | Regulation 16C (February 2025) | Already binding — act immediately | Binding now |
| Algo-ID implementation and white box/black box classification — every algo order must carry exchange-assigned Algo-ID; all AI/ML algos classified; black-box providers registered as Research Analysts | SEBI/HO/MIRSD/MIRSD-PoD/P/CIR/2025/0000013 | Mandatory from April 1, 2026 | Binding now |
| AI cybersecurity roadmap and SOC transformation plan — scenario-based AI risk testing, AI-augmented SOC plan, Zero Trust architecture, long-term agentic threat mitigation plan; IT committee ratification required | SEBI/HO/13/19/12(1)2026-ITD-1_CIMGI/10873/2026 | May 2026 circular — already binding | Binding now |
| Kill switch design for all AI/ML deployments — not just algo trading but all AI including fraud detection, customer service, recommendation AI; define activation triggers and authorization protocols before formal circular issues | August 2026 Pandey announcement (forthcoming circular) | Build now; circular expected Q4 2026–Q1 2027 | Forthcoming |
| Human oversight model mapping — define which HITL/HOTL/HIC model applies to each AI/ML deployment based on risk tier; document justification before formal circular defines requirements | August 2026 Pandey announcement (forthcoming circular) | Build now; map to anticipated tiered rules | Forthcoming |
| Vendor contract remediation — add audit rights, incident notification, DPDP Act warranties, performance SLAs to all vendor AI contracts under Regulation 16C accountability framework | Regulation 16C + August 2026 announcement | Immediately — Regulation 16C already binding | Binding now |
| Board-level AI governance framework — board-approved policy covering all five pillars of June 2025 consultation paper; document before formal circular requires it | June 2025 consultation paper (binding on adoption in final circular) | Build now; final circular will require it | Forthcoming |
SEBI, RBI, and CERT-In Are Converging on One AI Governance Architecture
Three patterns of convergence are visible in SEBI's evolving AI governance framework:
- Kill switch: CERT-In Blueprint (May 2026) for banking AI mandates kill switches at the model deployment level. SEBI's forthcoming circular announces the same requirement for securities market AI. Both frameworks were informed by the same regulatory concern: automated AI propagating adverse outcomes at machine speed across financial system participants before human intervention can stop it.
- Third-party accountability: RBI FREE-AI's Governance Pillar requires that banks retain accountability for vendor AI. Regulation 16C uses identical language for SEBI-regulated entities. MeitY's Seven Sutras Accountability Sutra establishes the same principle nationally. This is not coincidence — it is India's financial regulators adopting a unified position that the deploying entity, not the AI vendor, bears regulatory liability.
- Tiered proportionality: RBI FREE-AI scales obligations to bank AI portfolio size. IRDAI's working group mandate includes proportional capacity building. SEBI's forthcoming tiered rules follow the same architecture. For financial conglomerates that operate across banking (RBI), securities (SEBI), and insurance (IRDAI) — which describes every major Indian private sector bank and financial group — this convergence means a single AI governance program can satisfy all three regulators' frameworks with sector-specific adaptation rather than three separate compliance programs.
- DPDP Act as cross-sectoral data layer: All three regulators (RBI, SEBI, IRDAI) have now explicitly referenced DPDP Act 2023 alignment in their AI governance requirements. DPDP Act's non-discrimination, consent, and data minimization requirements apply to investor data in SEBI's perimeter, policyholder data in IRDAI's, and customer data in RBI's. A single DPDP-aligned data governance program for AI training and inference data satisfies the data governance pillar across all three regulators.