What AI governance obligations does SEBI's Responsible Usage of AI/ML Framework 2025 create for securities intermediaries?

SEBI's Guidelines for Responsible Usage of AI/ML in Indian Securities Markets (consultation paper June 20, 2025, followed by a September 2025 harmonized cross-exchange mechanism and an August 2026 final circular) create the first comprehensive cross-market AI/ML governance standard that applies simultaneously to ALL securities market intermediaries — stock exchanges, stockbrokers, asset management companies, investment advisers, clearing corporations, depositories, and other SEBI-registered entities. This framework is categorically distinct from SEBI's earlier AI-specific rules: the Algo-AI Guidelines (2012/2021) cover only algorithmic trading; the IA Second Amendment 2024 covers only investment advisers; the RA Third Amendment 2024 covers only research analysts. The Responsible Usage Framework applies to every AI/ML system that any SEBI intermediary uses for any purpose — trading, surveillance, risk management, client services, compliance, or operations. Six core obligations: (1) Non-delegable senior management accountability — senior management cannot outsource AI governance responsibility regardless of whether the AI tool was built in-house or procured from a third-party vendor; (2) Mandatory kill-switch provisions — all AI/ML systems must include an immediate halt capability; a system without a tested, functional kill-switch cannot be deployed in a SEBI-regulated context; (3) Mandatory human oversight at model deployment and monitoring stages — AI cannot operate in a fully autonomous mode without a designated human reviewer at key decision points; (4) Model governance infrastructure — skilled internal teams with AI/ML competency, not just IT teams; (5) Third-party vendor agreements with specific AI risk clauses — accountability for vendor AI does not transfer to the vendor; (6) Plain-language investor disclosures about AI use in services affecting their portfolios.

What is the kill-switch requirement under SEBI's AI/ML framework and how must intermediaries implement it?

SEBI's Responsible Usage of AI/ML Framework (2025) mandates that every AI/ML system deployed by a securities market intermediary include a functional, tested kill-switch with immediate halt capability. This is not a conceptual requirement — SEBI's examination teams expect intermediaries to demonstrate that the kill-switch works. Four implementation obligations: (1) Kill-switch existence and testing: every AI/ML system in production must have an operational kill-switch that a designated person can activate to immediately halt AI decision-making. The kill-switch must be documented, its functionality tested regularly, and test records maintained for SEBI examination; (2) Activation protocol: the circumstances under which the kill-switch must be activated must be specified in advance — not left to post-incident judgment. Typical triggers: AI system generating anomalous outputs, AI decisions conflicting with human review outcomes above a threshold, market conditions for which the AI was not trained, regulatory investigation; (3) Post-activation continuity: intermediaries must demonstrate they can continue core operations after a kill-switch activation — AI-dependent processes that have no manual fallback cannot satisfy the kill-switch requirement in spirit. If the AI cannot be switched off without halting client services, the kill-switch is illusory; (4) NSE enforcement layer: NSE Circular NSE/COMP/73568 (April 2026) requires all active trading members to submit half-yearly AI/ML usage reports to NSE including confirmation of kill-switch status for each AI/ML system in use. This creates a reporting cadence that makes non-compliance visible to both NSE and SEBI on a six-month cycle.

How does the SEBI AI/ML Framework 2025 differ from the existing SEBI Algo-AI Guidelines for algorithmic trading?

SEBI's Algorithmic Trading / AI-ML Guidelines (2012/2019/2021/2022 circulars) and the Responsible Usage of AI/ML Framework 2025 are parallel obligations with overlapping but non-identical scope. Understanding the difference is essential for compliance mapping: (1) Scope: Algo Guidelines cover algorithmic trading systems — systems that generate or execute buy/sell orders in the securities market at speed and scale. The Responsible Usage Framework covers ALL AI/ML systems used by intermediaries regardless of whether they touch trading. A stockbroker's AI-based customer onboarding system, an AMC's AI client risk profiler, or a clearing corporation's AI margin calculation model is outside the Algo Guidelines but inside the Responsible Usage Framework; (2) Primary obligation: Algo Guidelines govern the trading strategy AI — system approval, exchange co-location conditions, order-to-trade ratios, audit trails for trading algorithms. Responsible Usage Framework governs the governance architecture around any AI/ML deployment — accountability, oversight, kill-switch, vendor governance, disclosure; (3) Enforcement: Algo Guidelines are enforced through exchange-level system audits. The Responsible Usage Framework is enforced through SEBI's intermediary inspection framework, which covers all registered entities; (4) Interaction: a stockbroker's algorithmic trading AI is subject to both — it must comply with the technical requirements of the Algo Guidelines AND the governance requirements of the Responsible Usage Framework. The two sets of obligations do not replace each other. Brokers who have mapped compliance only to the Algo Guidelines have an uncovered gap under the Responsible Usage Framework for their algorithmic trading AI's governance architecture.

What third-party AI vendor governance obligations does SEBI's AI/ML Framework 2025 create for intermediaries using vendor AI platforms?

The majority of SEBI-registered intermediaries — particularly small and mid-size stockbrokers, regional AMCs, and independent investment advisers — use third-party AI platforms for client risk profiling, portfolio analytics, compliance monitoring, and fraud surveillance. SEBI's Responsible Usage Framework creates a senior management accountability obligation that does not transfer to third-party vendors: the intermediary is accountable for the AI, regardless of who built it. Four vendor governance obligations follow: (1) Pre-deployment AI assessment: before deploying a third-party AI tool, the intermediary must conduct an internal assessment of its compliance with SEBI's Responsible Usage Framework requirements — specifically kill-switch availability, human oversight integration points, auditability, and investor disclosure adequacy; (2) Vendor contract AI risk clauses: agreements with AI technology vendors must include specific risk provisions: model change notification (when the vendor retrains or updates the AI, the intermediary must know), AI system audit rights, data security and confidentiality obligations covering client data processed by vendor AI, and liability allocation for AI-generated errors; (3) Ongoing vendor AI monitoring: the intermediary cannot satisfy its accountability obligation through a one-time vendor assessment. Ongoing monitoring of vendor AI performance — for accuracy, discrimination, model drift, and regulatory compliance — is part of the senior management accountability obligation; (4) Vendor concentration risk: an intermediary where a single vendor's AI platform underpins multiple critical functions (trading surveillance, risk management, client onboarding) faces a concentration risk that the Responsible Usage Framework's governance principles require to be assessed and disclosed to the board.

AI Governance · SEBI · AI/ML Framework · Kill-Switch · All Intermediaries · India BFSI

SEBI Guidelines for Responsible Usage of AI/ML in Indian Securities Markets 2025: Kill-Switch Mandate, Human Oversight, Senior Management Accountability and Third-Party AI Vendor Governance for All Securities Intermediaries India

SEBI's Guidelines for Responsible Usage of AI/ML in Indian Securities Markets (consultation paper June 20, 2025; September 2025 harmonized mechanism; August 2026 final circular) create the first comprehensive cross-market AI/ML governance framework for ALL SEBI intermediaries — exchanges, stockbrokers, AMCs, investment advisers, clearing corporations, and depositories. Non-delegable senior management accountability, mandatory kill-switch for all AI/ML systems, human oversight at deployment and monitoring, AI-specific vendor contract clauses, and plain-language investor disclosures. NSE Circular NSE/COMP/73568 (April 2026) requires half-yearly AI/ML reporting from all trading members. Zero global competitor coverage.

Zytra Research Team · June 2025 · 12 min read

📖 Part of the India BFSI AI Governance Complete Guide 2026 — RBI, SEBI, IRDAI, NPCI & DPDP in one place.

India's securities market is one of the most AI-intensive in the world: algorithmic trading accounts for over 50% of exchange turnover, AI-driven risk surveillance runs across NSE and BSE in real time, and AI client onboarding and portfolio advisory operate across thousands of stockbrokers and AMCs. SEBI's existing AI rules were fragmented — Algo-AI Guidelines for trading, IA Amendment for advisers, RA Amendment for analysts. The Responsible Usage of AI/ML Framework 2025 closes the gap: a single, comprehensive AI governance standard that applies to all SEBI-registered entities, regardless of what their AI does.

The framework's two most operationally significant obligations — mandatory kill-switch and non-delegable senior management accountability — are not aspirational. NSE's April 2026 circular requiring half-yearly AI/ML reporting from every trading member creates an enforcement cadence that makes both obligations auditable on a six-month cycle. SEBI examination teams are expected to assess Responsible Usage Framework compliance as part of the standard intermediary inspection cycle beginning in FY 2027.

The Kill-Switch: From Concept to Compliance Evidence

SEBI's kill-switch requirement is the most operationally concrete AI governance obligation in the Responsible Usage Framework. A conceptual kill-switch — a policy that says "we can turn off the AI" without a tested, documented, regularly exercised mechanism — does not satisfy the requirement. Intermediaries preparing for SEBI examination need: per-system kill-switch documentation, activation protocol specifications, post-activation continuity procedures, and test records showing the kill-switch was exercised within the last six months. An AI system for which no kill-switch test record exists failed SEBI's standard before the examination began.

SEBI AI/ML Governance Checklist — All Intermediaries

☐ AI/ML system inventory: all AI/ML systems in use logged; categorized by function (trading, surveillance, risk, client services, compliance, operations); risk tiered
☐ Senior management accountability assignment: named senior manager accountable for each AI/ML system or domain; accountability non-delegable to vendor or IT team
☐ Kill-switch: each AI/ML system has a documented, tested kill-switch; activation protocol documented; post-activation continuity procedure in place; test records maintained (6-month cycle)
☐ Human oversight documentation: human review checkpoints at model deployment and monitoring stages documented per AI/ML system; reviewers named with qualifications
☐ Internal AI competency: skilled internal team or designated personnel with AI/ML competency identified and trained; not delegated entirely to IT or vendors
☐ Vendor AI governance: pre-deployment assessment documented; vendor contract AI risk clauses (change notification, audit rights, data security, liability); ongoing vendor AI monitoring
☐ Investor disclosure: plain-language disclosure of AI role in services affecting client portfolios; disclosure updated when AI's role changes materially
☐ NSE half-yearly reporting: AI/ML usage report filed with NSE per NSE/COMP/73568 (April 2026); kill-switch status confirmed per system; NIL filing if no AI/ML in use
☐ SEBI examination readiness: AI/ML inventory, accountability assignments, kill-switch test records, vendor assessment, investor disclosures — packaged for SEBI intermediary inspection