How does SEBI's February 2025 AI/ML Accountability Circular create sole accountability for Portfolio Managers using AI?

SEBI's Circular SEBI/HO/MIRSD-MIR/MIRSD-MIR/P/CIR/2025/9 (February 6, 2025) applied to all SEBI-regulated intermediaries — explicitly including Portfolio Managers registered under SEBI (Portfolio Managers) Regulations 2020 (SEBI/LAD-NRO/GN/2020/08, January 16, 2020). The February 2025 circular declared that SEBI intermediaries are 'solely responsible for the security, confidentiality, and integrity of investors' data' when using AI tools, and solely accountable for 'investment advice based on output of AI tools.' This sole accountability principle intersects with PMS-specific obligations in three ways: (1) No third-party AI vendor safe harbour: a Portfolio Manager that uses an external AI model for discretionary portfolio construction cannot attribute an adverse portfolio decision to the AI vendor. The PM is solely accountable for the AI's output — creating a model governance obligation where the PM must understand, validate, and monitor the AI it uses even when purchased from a third party; (2) AI disclosure in Disclosure Document: SEBI's PMS Regulations require Portfolio Managers to maintain and disclose their investment approach in a Disclosure Document provided to all clients. A PM using AI for discretionary portfolio decisions must now disclose this in the Disclosure Document — including what AI systems are used, for what functions, and the PM's governance over those systems; (3) Board-approved IT policy must address AI: PMS Regulations require a board-approved IT policy. Post-February 2025, this policy must explicitly address AI/ML system governance — not only general IT security.

What suitability AI governance obligations do SEBI PMS Regulations create for portfolio managers?

SEBI (Portfolio Managers) Regulations 2020 require Portfolio Managers to conduct a suitability assessment before onboarding any client, documenting the client's risk appetite, investment objectives, time horizon, and financial position. Portfolio Managers using AI for client risk profiling and suitability scoring — a common practice for robo-PMS platforms targeting clients at the Rs 50 lakh minimum ticket — face specific governance obligations under the suitability framework: (1) AI suitability score explainability: the suitability assessment must be documented and available for SEBI inspection; an AI suitability score that cannot be explained in terms of the client's disclosed financial parameters — if the AI's score is driven by features not disclosed in the client onboarding form — creates a documentation gap that fails SEBI's suitability standard; (2) Model drift and client relationship: suitability assessment is not a one-time exercise under PMS regulations; PMs must reassess suitability when client circumstances change. An AI suitability model that drifts — where the same client profile scores differently on the model at different times due to retraining — creates a compliance risk: the AI may generate different investment strategy recommendations for the same client at different intervals, inconsistent with the documented suitability assessment; (3) Concentration risk AI: SEBI PMS Regulations require PMs to avoid excessive concentration in individual securities or sectors unless appropriate for the client's risk profile. An AI portfolio construction model that optimizes for return while systematically underestimating concentration risk — a documented failure mode of AI portfolio optimizers trained on historical data from trending markets — creates a SEBI PMS compliance breach that cannot be attributed to the AI vendor under the February 2025 accountability circular.

What AI governance obligations arise from SEBI's Form BRA performance reporting requirement for Portfolio Managers?

SEBI requires Portfolio Managers to report performance in a standardized format (Form BRA — Balance Return Attribution) that must be provided to clients and disclosed on the PM's website. When AI drives portfolio construction and rebalancing decisions, Form BRA performance attribution becomes an AI governance question: how much of the return — positive or negative — came from AI-generated portfolio decisions versus fund manager judgment? Three specific AI governance implications for Form BRA: (1) AI vs human attribution: when a PM operates a mixed model (AI generates trade recommendations, fund manager approves or overrides), the performance attribution must be traceable to the decision source. This requires the AI system to log each recommendation and the fund manager's approval or override — creating an AI decision log that feeds Form BRA attribution; (2) Benchmarking AI strategies: SEBI requires PMs to benchmark their strategy performance against a disclosed benchmark index. AI strategies — particularly factor-based AI models — may generate performance patterns that diverge systematically from standard benchmarks during specific market regimes. Performance attribution that does not explain AI-driven factor exposures may mislead clients and SEBI inspectors; (3) Drawdown attribution: when AI portfolio construction produces a drawdown, the PM's ability to explain the drawdown to the client (required under PMS regulations) depends on the AI's explainability. A PM that tells a client 'the AI decided to hold these positions' without being able to explain why the AI held them fails the PM's disclosure obligation.

What AI governance gap exists for robo-PMS platforms at the SEBI minimum ticket size?

SEBI's minimum PMS ticket size is Rs 50 lakh per client. Robo-PMS platforms — digital portfolio management services using AI for fully automated discretionary portfolio construction — target HNI clients at or near this minimum ticket size, offering algorithmic investment strategies at lower fee points than traditional PMs. SEBI's PMS Regulations were written for human portfolio managers; robo-PMS creates a governance gap at the intersection of PMS obligations and AI automation: (1) No model validation standard for PMS: unlike registered investment advisers (who have a SEBI IA circular framework) or mutual funds (which have the 2019 AI/ML circular), Portfolio Managers have no SEBI circular addressing AI model validation, testing, or ongoing monitoring before or during deployment. A robo-PMS can deploy an AI portfolio construction model with SEBI's regulatory silence as the only governance framework; (2) Conflict of interest AI: PMS regulations prohibit conflicts of interest between the PM and the client. An AI portfolio construction model that is trained or optimized on the PM's own historical trading data may have embedded conflicts — learned behaviors that benefit the PM's order flow, prime brokerage relationships, or research provider relationships. No SEBI circular requires a robo-PMS to audit its AI model for embedded conflicts; (3) Algorithmic trading intersection: when a robo-PMS's portfolio AI generates orders that are executed through the PM's algorithmic trading system, the SEBI algorithmic trading framework (requiring exchange-approved algorithms) potentially applies — but SEBI has not defined whether a portfolio AI's output constitutes an 'algorithm' under the algorithmic trading circular or a 'discretionary decision' under the PMS framework. This jurisdictional gap leaves robo-PMS operating in regulatory ambiguity.

AI Governance · SEBI · Portfolio Management Services · PMS · Discretionary AI · India BFSI

SEBI Portfolio Management Services AI Governance: PMS Regulations 2020 Discretionary Portfolio AI Sole Accountability, Suitability AI Explainability, Form BRA Attribution and February 2025 AI Accountability Circular India

India's Portfolio Management Services industry manages Rs 30 lakh crore AUM across 400+ SEBI-registered Portfolio Managers. AI drives discretionary portfolio construction, client risk profiling, automated rebalancing, and performance attribution. SEBI's February 2025 AI/ML Circular made Portfolio Managers solely accountable for AI-generated investment decisions — no safe harbour for third-party AI vendors. SEBI PMS Regulations 2020 suitability documentation, Form BRA attribution, and IT policy obligations all intersect with this AI accountability. Zero governance framework defined. Zero global competitor coverage.

Zytra Research Team · February 2025 · 11 min read

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

India's Portfolio Management Services (PMS) industry — managing Rs 30 lakh crore in HNI client assets across 400+ SEBI-registered Portfolio Managers including IIFL Wealth, 360 ONE, Motilal Oswal PMS, Alchemy Capital, and Marcellus Investment Managers — entered a new AI accountability regime when SEBI's February 2025 Circular (SEBI/HO/MIRSD-MIR/MIRSD-MIR/P/CIR/2025/9) declared that all SEBI intermediaries bear sole responsibility for AI-generated investment decisions. Portfolio Managers using AI for discretionary portfolio construction, client suitability assessment, and rebalancing algorithms have no regulatory safe harbour for third-party AI vendor models.

The February 2025 circular's sole accountability principle intersects with the existing SEBI (Portfolio Managers) Regulations 2020 across four specific obligations: Disclosure Document AI disclosure, suitability documentation for AI-profiled clients, Form BRA performance attribution for AI-driven decisions, and board-approved IT policy AI governance. No SEBI circular has yet defined what AI governance looks like for each of these PMS-specific obligations.

Sole Accountability: What It Means for Third-Party PMS AI

Many Portfolio Managers license AI portfolio construction tools from external quantitative strategy providers, fintech platforms, or technology firms. Under the pre-February 2025 framework, a PM could argue that an AI vendor's model error contributed to a bad outcome. Under the February 2025 accountability circular, this argument is explicitly foreclosed: the PM is solely responsible for AI-generated investment decisions. The governance implication is direct: Portfolio Managers must now govern vendor AI as if it were internally built — validating it before use, monitoring it during use, and auditing it when outcomes are adverse.

Robo-PMS: The Regulatory Ambiguity Intersection

Robo-PMS platforms — fully automated discretionary portfolio management at the Rs 50 lakh minimum ticket — operate in a regulatory gap. SEBI's PMS Regulations apply but were written for human managers. SEBI's algorithmic trading framework may apply to AI-generated orders but has not been explicitly extended to portfolio-level AI. SEBI's February 2025 AI/ML circular applies to intermediaries but does not address PMS-specific compliance dimensions. The result: a robo-PMS can deploy a portfolio AI without any SEBI-mandated validation standard, model testing requirement, or minimum performance threshold.

PMS AI Governance Checklist

☐ AI disclosure in Disclosure Document: AI systems used for discretionary decisions disclosed to clients; functions (portfolio construction, rebalancing, suitability) named; PM governance over vendor AI described; SEBI February 2025 accountability framework referenced
☐ Suitability AI explainability: AI risk profiling scores explainable in terms of disclosed client parameters; suitability documentation does not rely on unexplained AI outputs; model drift monitoring for consistent suitability scoring over client relationship
☐ Third-party AI vendor governance: vendor AI validation conducted before use; model change notification contractual right; performance monitoring cadence; sole accountability compliance documented per February 2025 circular
☐ Form BRA AI attribution: AI decision vs fund manager override log feeds performance attribution; AI factor exposures documented for benchmark divergence explanation; drawdown AI attribution explainable to client
☐ Board IT policy AI update: board-approved IT policy updated to cover AI/ML system governance; AI model inventory, validation, monitoring obligations included; SEBI February 2025 circular compliance documented
☐ Concentration risk AI: AI portfolio optimizer tested for concentration risk underestimation in trending markets; SEBI PMS concentration limits compliance monitoring independent of AI model output
☐ SEBI examination readiness: Disclosure Document AI section, suitability AI audit trail, vendor AI governance evidence, Form BRA AI attribution documentation — packaged for SEBI PMS inspection