SEBI Depository Participant AI Governance 2026: NSDL, CDSL, and DP Obligations

India's depository infrastructure is the backbone of the securities market — 17+ crore demat accounts holding ₹350+ lakh crore in assets. NSDL and CDSL, and the 900+ depository participants registered with them, have deployed AI across the account lifecycle: at opening (eKYC, identity verification), during holding (fraud detection, beneficial ownership tracing), and at pledging (automated pledge creation, invocation, and release). SEBI CSCRF (effective January 2025) and the 2024 cybersecurity circular create specific governance obligations for this AI — but no analysis maps these obligations to the depository and DP context specifically.

This analysis does that mapping. The governance obligations differ materially between depositories (classified as market infrastructure institutions under CSCRF's highest tier) and their DPs (tiered by size and risk), and the AI use cases in depositories — demat fraud detection, pledge management AI, PMLA beneficial ownership tracing — carry governance requirements that generic AI governance frameworks do not address.

India's Depository AI Landscape

Infrastructure by Numbers (2026)

Six Primary AI Use Cases

1. Account validation AI: verification of PAN-Aadhaar linkage, address validation, identity verification for new demat account opening — AI is the primary processing layer for the millions of new accounts opened monthly.
2. Beneficial ownership tracing: AI graph analysis to trace ultimate beneficial ownership through holding structures — relevant for PMLA compliance and SEBI surveillance. Complex corporate structures require AI traversal that manual processes cannot scale to.
3. Pledge and lien management AI: automated processing and validation of pledge creation, invocation, and release. Critical for margin trading, ESOP management, and NBFC lending against securities — high volume, time-sensitive, high financial consequence.
4. Demat fraud detection: AI detection of unauthorized dematerialization, forged signatures, and fake share certificates. Fraud at scale is only detectable through AI pattern recognition — manual review cannot cover the volume.
5. KRA (KYC Registration Agency) AI: KYC data validation and risk scoring for DP onboarding of new clients. AI processes KYC records from multiple KRAs and flags inconsistencies or risk indicators.
6. Corporate action processing AI: automated processing of dividend, bonus, rights entitlements and reconciliation. Errors in corporate action AI affect large numbers of retail investors simultaneously.

Regulatory Framework: What Applies to Depositories and DPs

SEBI CSCRF (effective January 2025)

SEBI's Cybersecurity and Cyber Resilience Framework classifies market infrastructure institutions (MIIs) — which includes NSDL and CDSL as depositories — at the highest tier with the most stringent cybersecurity requirements. DPs are classified at lower tiers based on size and risk profile but remain in scope.

Key CSCRF AI implications for depositories and DPs:

• AI systems used in critical functions (settlement, custody, clearing) are "critical IT assets" under CSCRF — subject to annual VAPT, penetration testing, and real-time incident reporting to SEBI.
• AI models that affect settlement or custody outcomes must be documented in the technology risk register maintained under CSCRF.
• Algorithmic systems (including AI) must have governance controls, change management procedures, and comprehensive audit trails — a regulatory documentation obligation, not merely a best practice.

SEBI Cybersecurity Circular 2024

The 2024 cybersecurity circular updated obligations for market participants including DPs. AI-specific provisions:

• Third-party AI providers used by DPs are covered under CSCRF third-party risk management — the DP is responsible for ensuring its AI vendor meets SEBI cybersecurity standards. Outsourcing AI does not outsource the regulatory obligation.
• AI-assisted KYC and account validation must maintain audit trails sufficient for regulatory examination. A SEBI inspection that identifies a customer account validation error must be traceable back to the AI decision that drove it.
• Data processed by AI must comply with SEBI's data localization and storage requirements. AI models trained on depository data are subject to these requirements at training and inference.

PMLA/FIU AI Obligations

Depositories are reporting entities under PMLA. AI used for suspicious transaction reporting (STR) generation must be governed under PMLA compliance — an AI model's decision to flag or not flag a transaction for STR is a regulatory decision with direct legal implications. A missed STR resulting from an AI false negative creates PMLA liability for the depository or DP, not for the AI vendor.

Demat Fraud Detection AI: Specific Governance Requirements

Demat fraud is a significant regulatory concern — fraudulent dematerialization, pledge fraud, and unauthorized account access affect retail investors at scale. AI is the primary defense layer, and AI governance determines the effectiveness of that defense.

Forgery Detection AI

AI that detects forged physical share certificate submissions for dematerialization must be validated against ground truth: what percentage of forged certificates does it catch? What is the false rejection rate for legitimate certificates? CSCRF requires performance documentation for security-critical AI systems. A forgery detection AI that rejects 5% of legitimate certificates creates a significant customer harm — investors unable to dematerialize valid physical holdings.

Suspicious Account Activity AI

Pattern detection for unauthorized access attempts, unusual pledge creation/invocation patterns, and account takeover indicators is MRMF-adjacent — a SEBI equivalent of model performance monitoring for market infrastructure AI. Key metrics that must be monitored: detection rate for known attack patterns, false positive rate (accounts incorrectly flagged), and time-to-detection for novel attack vectors. CSCRF's incident reporting requirement means that when suspicious activity AI fails to detect a fraud event that results in a reportable incident, the AI's performance gap becomes a supervisory data point.

Beneficial Ownership Graph AI

AI that traverses corporate holding structures to identify ultimate beneficial owners for PMLA compliance is high-stakes — a missed beneficial owner creates regulatory liability. Validation must include complex corporate structure test cases: multi-layer holding companies, trust structures, and NRI holding patterns that are common in India's capital markets. A beneficial ownership AI validated only against simple corporate structures will fail on the complex structures where PMLA enforcement matters most.

PAN-Aadhaar Linkage Validation AI

With SEBI's mandatory PAN-Aadhaar linkage for demat accounts, AI that validates this linkage must achieve near-zero false rejection rate for legitimate linked accounts. False rejections block investor access to their securities holdings — a direct customer harm with regulatory attention. The governance obligation: validation against a representative sample of linked account pairs, including edge cases (name mismatch in PAN vs Aadhaar, recently linked accounts), before production deployment.

Pledge Management AI: Governance for Margin and ESOP Use Cases

Pledge and lien management is one of the highest-value AI applications in depositories. Margin trading, ESOP management, and NBFC lending against securities all depend on accurate, timely pledge processing. Errors here create immediate financial harm.

Margin Pledge AI (Post-2020 SEBI Circular)

SEBI's 2020 circular on margin obligations requires client pledges to be created on exchange-approved depositories. AI that processes pledge creation, invocation, and release must meet three governance requirements under CSCRF:

• Auditable: every AI-assisted pledge action must generate a complete, tamper-evident audit log including input parameters, model version, output (action taken), timestamp, and operator who approved. Required for SEBI examination and dispute resolution — when a client disputes a pledge invocation, the audit log is the evidentiary record.
• Accurate: pledge invocation errors (incorrect shares invoked, wrong quantity) create immediate financial harm and regulatory exposure. AI accuracy must be validated before deployment and monitored continuously — a drift in pledge accuracy is a material AI risk event.
• Latency-bounded: margin calls require pledge invocation within prescribed settlement timelines. AI processing latency must be within SEBI's settlement timelines — a technically sound AI model that is too slow to process pledge invocations under peak load fails its governance requirement.

ESOP Pledge Management

Employee stock option pledges are complex — vesting schedules, exercise prices, tax implications at multiple stages. AI managing ESOP pledges must handle these complexities accurately. Governance should include: a test suite for vesting calculation accuracy covering standard and accelerated vesting schedules, error rate monitoring in production, and an exception queue for AI-uncertain decisions that routes to human review before the pledge action is executed.

NBFC Securities Lending AI

NBFCs using depositories for securities-based lending use AI to value collateral and determine lending limits. This AI is MRMF-adjacent — if the NBFC is regulated by RBI, the credit decision AI (including collateral valuation) falls under RBI MRMF. For bank-affiliated DPs facilitating NBFC securities lending, both SEBI CSCRF (for the depository function) and RBI MRMF (for the credit decision) apply. Governance complexity is highest at this intersection.

DP-Specific AI Governance: The 900+ Entity Challenge

The 900+ registered DPs range from major banks (SBI, HDFC, Kotak — with sophisticated AI governance programs) to small regional brokers with minimal AI governance capability. SEBI faces a significant regulatory challenge in achieving consistent AI governance across this range. The governance framework must work for both ends of the spectrum.

Risk-Tiered Approach

CSCRF already tiers DPs by size and risk — this tiering should extend to AI governance requirements. Large DPs (bank-affiliated, large broker networks with millions of accounts) should face MRMF-equivalent AI governance requirements: formal model inventory, independent validation, performance monitoring frameworks, and documented change management. Small DPs with limited AI deployment — primarily using third-party platforms with AI components they do not develop or configure — should face simplified documentation requirements: vendor AI disclosure, due diligence evidence, and incident notification procedures.

DP Third-Party AI

Most smaller DPs do not build their own AI — they use third-party KYC platforms, account management systems, and analytics tools that include AI components. SEBI CSCRF third-party risk management applies: the DP is responsible for the AI governance of its technology vendors. The practical obligation is not to validate the vendor's AI model (which the DP typically cannot do) but to: obtain vendor AI documentation, include audit rights in contracts, require vendor compliance certifications equivalent to CSCRF requirements, and establish incident notification obligations.

Account Opening AI (eKYC)

Digital demat account opening uses AI for video KYC, document verification, and liveness detection. These AI systems must comply with SEBI's KYC circulars and PMLA requirements simultaneously. The specific governance obligation that is most frequently missed: AI rejections of legitimate customer applications must be reviewable by a human within a prescribed timeline. An eKYC AI that incorrectly rejects a legitimate application and routes it to a queue with a 14-day review cycle creates a customer harm that SEBI's investor protection mandate addresses directly.

Nominee Validation AI

With SEBI's enhanced nominee registration requirements, AI processing nominee verification must achieve high accuracy. Errors in nominee registration directly affect inheritance and transmission of securities — a harm that typically becomes visible only at the worst possible time. Governance should include: test coverage for standard and complex nominee scenarios (multiple nominees, minor nominees, NRI nominees), false rejection rate monitoring, and clear human escalation for AI-uncertain cases.

Five AI Governance Controls for Depositories and DPs

1. Critical AI Asset Register

Identify all AI systems that touch settlement, custody, pledge, and fraud detection functions. Register as critical IT assets under CSCRF. For each system, the register must include: model version, vendor or in-house status, deployment date, last validation date and outcome, incident history, and next scheduled review. This register is the foundational document for SEBI examination of AI governance — without it, no other governance activity is demonstrable at examination time.

2. Pledge AI Audit Trail

Every AI-assisted pledge action (creation, invocation, release) must generate a complete, tamper-evident audit log. Required fields: input parameters (account, securities, quantity, trigger), model version used, output (action taken and parameters), timestamp, and identity of the operator who confirmed or overrode the AI recommendation. This log serves two purposes: SEBI examination (regulatory requirement under CSCRF) and dispute resolution (evidentiary record when clients dispute pledge actions).

3. Beneficial Ownership AI Validation

Conduct quarterly validation of beneficial ownership tracing AI against known test structures. Where possible, use regulatory test cases from SEBI's surveillance team; otherwise construct test cases covering: pyramid holding companies (common in promoter holding structures), family trust structures (common for HNI retail investors), and NRI holding through FPI structures (common in institutional holding). Document the AI's performance on these test cases and the false-negative rate for complex structures where PMLA enforcement risk is highest.

4. Third-Party AI Due Diligence

Conduct annual review of all third-party AI providers (eKYC platforms, fraud detection services, analytics tools) against CSCRF third-party risk management requirements. The due diligence package should include: AI model documentation review (what does the vendor's AI do and how is it validated?), vendor security certifications, vendor incident history for the AI system, data processing agreement review for alignment with SEBI data requirements, and contract clause verification (audit rights, incident notification timelines, change notification requirements).

5. Retail Investor Impact Monitoring

AI systems that affect retail investor account access (eKYC rejections, account freezes, pledge invocation) must be monitored for false rejection rates. Produce a monthly report covering: eKYC rejection rate vs. baseline (to detect drift in AI performance), account freeze rate (to detect AI over-flagging), and pledge error rate (incorrect invocations or releases). Threshold breaches should trigger an immediate investigation — a sudden increase in eKYC rejections is more likely to indicate AI model drift than an increase in fraudulent applications.

Frequently Asked Questions