What is SEBI's white-box vs black-box AI classification for retail algo trading and what governance obligations does it create?

SEBI Circular SEBI/HO/MIRSD/MIRSD-PoD/P/CIR/2025/0000013 (February 4, 2025, fully effective April 1, 2026) introduces India's first regulatory classification of AI trading algorithms by their transparency characteristics — the white-box/black-box distinction. The classification creates fundamentally different regulatory pathways for transparent versus opaque AI: (1) White-box AI algos: algorithmic trading systems whose trading logic is transparent, replicable, and can be fully documented — a rule-based momentum strategy, a simple technical indicator system, or any algorithm whose decision logic can be described in full and independently verified. White-box algos qualify for a fast-track exchange registration process. The exchange can verify that the documented logic matches the algo's actual behaviour, enabling streamlined approval without SEBI Research Analyst registration; (2) Black-box AI algos: algorithmic trading systems whose logic is non-replicable or non-transparent — machine learning models, neural networks, or any system where the trading signal cannot be derived from a readable rule set. For black-box AI algos, the circular mandates that the algo provider must register as a SEBI Research Analyst. The RA registration creates two major obligations: the RA must maintain a detailed research report for each AI algo explaining its strategy, risk parameters, and methodology to the extent possible; and any material change in the AI model's logic requires re-registration as an entirely new algorithm. The white/black-box classification is India's operational implementation of AI explainability in a high-stakes financial context — it is the only regulatory framework globally that directly links AI opacity to a different regulatory approval pathway.

What broker principal liability obligation does SEBI's retail algo trading circular create for AI-generated trading orders?

The broker principal liability provision in SEBI/HO/MIRSD/MIRSD-PoD/P/CIR/2025/0000013 is among the most commercially significant AI governance obligations for India's stockbroking sector. Three key liability provisions: (1) Broker as principal for AI algo outputs: stockbrokers that provide API access to retail investors for algorithmic trading bear full liability for AI algo outputs to their retail clients. The broker cannot disclaim liability by pointing to the third-party algo provider — whether the algo was developed in-house, licensed from an algo provider, or operated through a co-branded arrangement, the broker is the principal accountable party for investor harm arising from AI algo malfunctions, errors, or market abuse; (2) Unique identifier traceability: all AI-generated orders must carry a unique identifier that traces the order to the specific algorithm that generated it. This creates an AI order attribution requirement — every trade executed by an AI algo must be identifiable as AI-generated and attributable to a specific registered algo ID. Unattributed AI orders (where the broker cannot identify which algorithm generated a trade) are non-compliant; (3) Algo provider as broker agent: when a broker uses an external algo provider's system, that provider operates as the broker's agent for regulatory purposes. The broker's principal liability extends to the external provider's AI system behaviour. Contracts with algo providers that purport to limit the broker's liability for provider AI failures are commercially ineffective against SEBI's regulatory accountability framework. The practical implication: brokers offering third-party AI trading systems through their platforms (a common model in India's growing retail algo trading market) carry regulatory liability for the AI behaviour of systems they did not build — creating strong incentives for broker-level AI governance of algo providers.

What are the exchange-level kill-switch and simulation testing obligations for AI trading algorithms under SEBI's 2025 circular?

SEBI's retail algo trading circular creates two exchange-level AI governance obligations that give NSE, BSE, and MCX direct authority over AI trading algorithms: (1) Exchange-level kill-switch: stock exchanges must maintain the operational capability to immediately halt all orders from a specific AI algo ID. The kill-switch must operate at the algo ID level — not at the broker level or the investor level. This means the exchange must be able to surgically stop one AI algorithm's orders while other algorithms from the same broker continue to operate. For exchanges, this requires infrastructure that tracks the AI algo ID attached to every order in real time and can selectively block by ID within milliseconds. For brokers, it means every AI algo must have a registered ID that the exchange can reference and halt on demand; (2) Simulation testing: exchanges must conduct surveillance on AI algo behaviour including simulation testing. Simulation testing requires the exchange to run each registered AI algo against historical market data to verify that the algo behaves as described in its registration documentation — that it does not exhibit undisclosed market manipulation patterns, excessive cancellation behaviour, or spoofing-adjacent strategies that are invisible in live order flow but visible in simulation. For black-box AI algos where the algo provider has registered as an RA but the ML model's logic cannot be fully documented, simulation testing is the exchange's primary verification mechanism. Together, the kill-switch and simulation testing obligations create exchange-level AI governance infrastructure that goes beyond regulatory oversight — the exchange is actively monitoring, testing, and controlling AI trading system behaviour in real time.

How does SEBI's retail algo trading AI classification interact with India's broader AI governance framework for securities markets?

The SEBI retail algo trading circular (February 2025) sits within a multi-layer AI governance stack for India's securities markets: (1) Interaction with SEBI AI/ML Responsible Usage Framework 2025: the Responsible Usage Guidelines (June 2025 consultation, August 2026 final) apply to all SEBI intermediaries including stockbrokers and algo providers. The Responsible Usage Framework's kill-switch, human oversight, and senior management accountability provisions apply to AI trading systems — the retail algo circular's exchange-level kill-switch is a sector-specific implementation of the Responsible Usage Framework's general kill-switch principle; (2) Interaction with SEBI RA Amendment 2024: the requirement that black-box AI algo providers register as SEBI Research Analysts directly links to the SEBI Research Analysts Amendment 2024 (already covered in Zytra's corpus), which mandates disclosure of AI use in research and requires full RA accountability for AI-generated outputs. An algo provider that registers as an RA for its black-box AI trading system comes under the full RA accountability framework — including the obligation to maintain audit trails for AI model inputs and outputs; (3) Interaction with SEBI Stockbrokers Institutional Mechanism 2024: the Stockbrokers Institutional Mechanism circular (SEBI/HO/MIRSD/MIRSD-PoD-1/P/CIR/2024/96, July 2024) mandates AI trading surveillance at the broker level for detecting market abuse. The retail algo trading circular complements this by requiring AI trading systems to carry unique identifiers — making the broker's AI surveillance system able to identify which algorithm generated which orders when reviewing for market abuse patterns; (4) Unique India-specific AI governance contribution: the white-box/black-box classification is an India-specific regulatory innovation with no direct equivalent in EU AI Act, SEC market structure rules, or UK FCA algorithmic trading guidance. It operationalizes AI explainability as a binary regulatory classification rather than a principles-based obligation — a pragmatic approach to AI model transparency in high-frequency, high-stakes financial contexts.

AI Governance · SEBI · Algo Trading · AI Classification · Broker Liability · India BFSI

SEBI Retail Algo Trading AI Governance Framework 2025: White-Box vs Black-Box AI Algorithm Classification, Mandatory SEBI Research Analyst Registration for Black-Box AI Algos, Broker Principal Liability and Exchange Kill-Switch for 1,500+ Stockbrokers India

SEBI Circular SEBI/HO/MIRSD/MIRSD-PoD/P/CIR/2025/0000013 (February 4, 2025, effective April 1, 2026) introduces India's first regulatory classification of AI trading algorithms by transparency: white-box AI algos (transparent, replicable logic) qualify for fast-track exchange registration; black-box AI algos (non-replicable ML models) must register algo providers as SEBI Research Analysts with detailed research reports per AI model — any change in AI logic requires re-registration as a new algorithm. Stockbrokers bear full principal liability for AI algo outputs to retail investors. All AI orders must carry unique identifiers traceable to the generating algorithm. Stock exchanges must maintain algo-ID-level kill-switches and conduct simulation testing of registered AI algos. Applies to all ~1,500 stockbrokers, all algo providers, NSE/BSE/MCX as oversight bodies. Zero global competitor coverage.

Zytra Research Team · February 2025 · 12 min read

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

The white-box/black-box AI classification in SEBI's February 2025 circular is one of the most practically significant AI governance innovations in any global financial regulation. Most AI governance frameworks approach model transparency as a principles obligation — "be explainable," "document your model." SEBI converts that principle into a binary regulatory classification that directly determines the regulatory pathway: transparent AI gets fast-track approval; opaque AI must acquire SEBI Research Analyst registration with ongoing documentation and re-registration obligations for every model change.

For India's rapidly growing retail algo trading market — where thousands of retail investors deploy AI trading strategies through broker APIs — the broker principal liability provision is commercially transformative. A broker that offers a third-party AI trading system to retail clients carries full regulatory liability for that AI's outputs. The incentive to deeply understand and govern every AI system on its platform is now a regulatory obligation, not just business practice.

The Re-Registration Obligation for Black-Box AI

The requirement to re-register black-box AI algos as new algorithms whenever the AI model logic changes creates a significant operational constraint on ML-based trading systems. ML models evolve continuously — retraining on new data changes model parameters and output behaviour. Under SEBI's framework, a meaningful retrain that changes the model's logic effectively creates a new algorithm requiring a new RA research report and exchange registration. This makes rapid iterative model development incompatible with the black-box registration pathway — algo providers face a choice between moving to white-box logic (replicable rules) or accepting slow, documentation-heavy update cycles for black-box ML systems.

SEBI Retail Algo Trading AI Compliance Checklist

☐ White/black-box classification: all trading algorithms classified; documentation of classification rationale; white-box algos have full logic documentation; black-box algos identified for RA registration pathway
☐ RA registration for black-box algos: algo providers of black-box AI systems registered as SEBI Research Analysts; research report per AI algo maintained; re-registration process defined for material model changes
☐ Exchange registration: all algos registered with relevant exchange (NSE/BSE/MCX); unique algo ID assigned; fast-track or standard pathway per white/black-box classification
☐ Unique order identifier: all AI-generated orders carry unique identifier traceable to registered algo ID; order attribution auditable
☐ Broker principal liability documentation: broker accountability for algo outputs documented; algo provider contracts reviewed for liability alignment; AI system due diligence for third-party algo providers
☐ Security controls: OAuth authentication, static IP whitelisting, 2FA on all API channels for AI algo operation
☐ Exchange kill-switch readiness: kill-switch mechanism operational for algo-ID-level halt; tested; procedures for kill-switch invocation defined
☐ SEBI examination readiness: classification documentation, RA registration records, exchange registration, order attribution logs — packaged for SEBI inspection