Skip to main content

SEBI AI/ML Compliance

SEBI has layered its AI/ML regulation over three instruments: the 2019 reporting circulars for market intermediaries, Regulation 16C (introduced by the SEBI (Intermediaries) (Amendment) Regulations, 2025 on 10 February 2025), and the 20 June 2025 Consultation Paper on guidelines for responsible AI/ML usage in Indian securities markets. Combined, these set requirements for AI systems used in securities markets, trading workflows, asset management platforms, and regulated financial services in India. Check SEBI's circulars page for the current status of the Consultation Paper's downstream rules.

The three SEBI instruments that apply today​

1. The 2019 reporting circulars (live, in force)​

  • SEBI/HO/MIRSD/DOS2/CIR/P/2019/10 (4 January 2019) — stock brokers + depository participants
  • SEBI/HO/MRD/DOP1/CIR/P/2019/24 (31 January 2019) — market infrastructure institutions (exchanges, clearing corporations, depositories)
  • SEBI/HO/IMD/DF5/CIR/P/2019/63 (9 May 2019) — mutual funds, AMCs, trustee companies

All regulated intermediaries must file AI/ML-system disclosures — including a NIL declaration when no system is deployed — via the Exchange's Enhanced Supervision Portal. Reporting is half-yearly for trading members (April-September and October-March windows). The filing captures name + type of AI/ML application, date of first use, whether it initiates / routes / executes orders, whether it falls under discretionary investment or portfolio management, use in cybersecurity, marketing claims about the system, safeguards against abnormal behaviour, implementation methodology, and compliance attestation.

Harmonisation since July 2025: NSE-registered trading members file a single disclosure that NSE shares with BSE, MSE, MCX, NCDEX. (MSE adopted the harmonised model formally in MSE Circular MSE/MEM/17921/2025 on 7 October 2025.)

2. Regulation 16C — binding sole-liability rule (in force 10 February 2025)​

SEBI Board's 208th meeting notified three amendment regulations: SEBI (Intermediaries) (Amendment) Regulations, 2025 (inserts Regulation 16C), plus parallel amendments on the MIIs and Depositories side (the DP Amendment Regulations came into force 1 April 2025).

Regulation 16C's operative rule: any SEBI-regulated entity is solely liable for AI/ML tools it uses, whether developed in-house or procured externally. Liability covers data privacy and security, integrity of AI outputs, and compliance with all applicable laws. Non-compliance triggers SEBI Act 1992 enforcement — monetary sanctions, cease-and-desist, suspension.

3. June 2025 Consultation Paper​

Released 20 June 2025; comments closed 11 July 2025. Proposes a board-approved AI Governance Framework for all regulated institutions with internal controls, algorithmic audits, model explainability, fallback plans, continuous monitoring, independent audits, periodic accuracy reporting to SEBI, investor-protection disclosures, regulatory sandbox testing, staff training, and human-in-the-loop mechanisms. Any downstream rules or notifications are published on SEBI's circulars page.

A concrete example: a securities-firm research assistant​

Here is how a SEBI-regulated research house lands on AxonFlow. They want to deploy an AI research assistant that pulls from internal reports, public filings, and SEBI-registered data sources, and drafts analyst notes.

What SEBI asks the firm to prove:

  • 2019 circular disclosure: the assistant is disclosed via the half-yearly ESP filing — covering name, date of first use, implementation methodology, safeguards, and a compliance attestation.
  • Regulation 16C liability: the firm is solely liable for outputs, even though the underlying model is a third-party foundation model. The SLA with the vendor does not transfer SEBI liability.
  • DPDP Act 2023 alignment: when the assistant touches client personal data (contact details, PAN for KYC cross-reference), it must honour DPDP rules once Phase 3 kicks in on 13 May 2027.
  • Requirements proposed in the June 2025 Consultation Paper: board-approved governance framework, model explainability, fallback plan, continuous monitoring, independent audit, HITL on material analyst notes.

How it maps to AxonFlow tiers:

NeedCommunityEvaluationEnterprise
PAN / Aadhaar / Indian-phone detection on all inbound connector dataSystem policies (sys_pii_pan, sys_pii_aadhaar, plus platform Indian detectors)SameSame + checksum-aware validation
Policy enforcement on every LLM + MCP call (Regulation 16C "integrity of AI outputs")YesYesYes
Full audit trail with decision chain and transparency headers (Consultation Paper explainability)YesSameSame + 10-year retention (AuditRetentionDays=3650)
HITL on analyst-note drafts before they land with clients (Consultation Paper HITL)Can emit require_approval decisions; no queue to act on themHITL approval queue (24h expiry, 100 pending cap)Production HITL queue + portal
Policy simulation before a DPDP-driven policy change hits productionNot providedPolicy simulation (Evaluation-tier limits)Simulation + production reporting
Periodic audit export for the 2019-circular half-yearly filingRaw audit log; team builds the export itselfSamePOST /api/v1/sebi/audit/export + GET /api/v1/sebi/audit/export/{id} — structured export designed to be filing-ready
Retention-window proof (for DPDP + SEBI data-handling audit)Tier-level audit retention, limitedSame, extended windowGET /api/v1/sebi/audit/retention surfacing the actual retention state
SEBI-readiness dashboard ("where do we stand on our half-yearly filing?")Not providedNot providedGET /api/v1/sebi/audit/readiness + GET /api/v1/sebi/dashboard

What the 5 Enterprise SEBI endpoints do in practice​

The SEBI module surface is intentionally small — 5 endpoints — because SEBI's own expectations around AI/ML are narrower than, say, RBI FREE-AI's 26 recommendations. What matters is that each endpoint maps to an actual compliance workflow rather than a generic "export".

  • POST /api/v1/sebi/audit/export — create an audit export job for a date range + scope (tenant, connector, user). Designed to produce the structured evidence a compliance officer submits alongside the half-yearly ESP filing.
  • GET /api/v1/sebi/audit/export/{id} — poll an export job. Returns manifest metadata (tenant IDs, row counts, time bounds) the compliance officer sees before sharing with the internal auditor.
  • GET /api/v1/sebi/audit/retention — confirm retention configuration on the tenant: how long audit data is kept, where it lives, how it survives tenant deletion. Comes up during SEBI TRM inspections.
  • GET /api/v1/sebi/audit/readiness — a readiness summary. Answers "are we ready to file?" before the compliance deadline: missing metadata, gaps in attestations, expired policies, etc.
  • GET /api/v1/sebi/dashboard — the aggregated operational view across recent exports, readiness, and compliance posture. What the head of compliance looks at monthly.
Decision-chain reconstruction (v8.7.0)

As of platform v8.7.0 the SEBI audit export is derived from the canonical audit_logs decision rows and groups them by correlation_id (the W3C trace ID a Policy Enforcement Point propagates across a request's hops). The export returns reconstructed decision chains under data.decision_chains, so a Regulation 16C reconstruction — "what did the model see, what did it produce, who approved it" — comes back as one ordered chain per logical request rather than a flat chronological dump. See Audit Logging → decision chains.

For most securities firms, these 5 endpoints cover the SEBI workflow. Additional reporting needs (e.g., model inventory beyond the ESP filing) are usually handled through the broader governance surface, not a SEBI-specific endpoint.

For the cross-framework view of regulated API families, use the Enterprise Compliance API Surface.

DPDP Act 2023 — the compliance clock​

The Digital Personal Data Protection Act 2023 was assented to on 11 August 2023. The DPDP Rules 2025 were notified on 13 November 2025. Rollout is phased:

  • Phase 1 (14 November 2025 — live now): procedural provisions — Data Protection Board constitution, definitions, bar of civil-court jurisdiction. No substantive compliance obligations on firms yet.
  • Phase 2 (13 November 2026): Consent Manager registration process operationalised.
  • Phase 3 (13 May 2027): substantive compliance goes live — grounds for processing, notice + consent, rights of data principals, additional obligations of Significant Data Fiduciaries, cross-border transfer rules, data breach notifications, processing of children's personal data.

What engineers should be doing now, not in 2027:

  • PII detection + redaction on every prompt, response, and connector result touching Indian data subjects. AxonFlow's sys_pii_pan and sys_pii_aadhaar system policies are the baseline.
  • Audit trail with enough granularity that a Data Principal's right-to-access request (activated Phase 3) can be answered without a database archaeology expedition.
  • Policy hooks for purpose-limited processing — defined now even if only logged today, because the DPDP notice regime requires they be enforceable by May 2027.

Non-compliance penalties in the DPDP schedule are not theoretical: up to ₹250 crore for failure to maintain reasonable security safeguards, up to ₹200 crore for breach-notification failures, up to ₹150 crore for Significant Data Fiduciary obligation breaches.

What Community covers​

Community is a credible starting point for SEBI-relevant engineering work:

  • audit logging
  • system and tenant policy enforcement
  • India-relevant PII protection (sys_pii_pan, sys_pii_aadhaar, plus the broader Indian-detector set covering UPI, IFSC, bank-account, GSTIN, voter ID, driving licence, passport, Indian phone, pincode, ration card — 12 detectors total on the india_pii_detector module)
  • governed LLM and MCP execution paths that are easy to review

This is enough for engineering teams building research assistants, advisory copilots, analyst tools, trading-support workflows, or internal AI operations tools — before they touch the Enterprise SEBI module.

Industry playbook​

Research-assistant AI​

The concrete flow above. Community covers policy + audit; Evaluation adds the HITL queue for analyst-note review; Enterprise adds the filing-ready export and readiness dashboard. Regulation 16C sole-liability is why the audit trail matters: when a recommendation turns out to be wrong, the firm must be able to reconstruct what the model saw, what it produced, and who approved it.

Advisory copilot​

For investment-advisory workflows, the DPDP Phase 3 clock matters most: advisory conversations often carry PAN, account numbers, and client-specific recommendations. AxonFlow's Indian-PII detectors plus tenant policies enforcing purpose-limited disclosure let the advisory team build now and stay compliant when Phase 3 activates.

Fraud and surveillance AI​

Exchange-side surveillance uses AI to flag anomalies. The 2019 circular requires reporting; Regulation 16C makes the exchange liable for outputs; the June 2025 Consultation Paper adds accuracy reporting to SEBI. AxonFlow's audit surface is often the cleanest way to produce both the half-yearly ESP filing and the accuracy evidence the Consultation Paper would require.

Engineering perspective​

For staff-level engineers at an Indian securities firm: Community is a real foundation, not a tease. You can build and run a governed AI product end-to-end on Community, prove controls with real Indian data, and stay aligned with the 2019 circular filing workflow. The Enterprise SEBI module becomes the cleaner operating model the moment your compliance team wants structured exports, a readiness dashboard, or 10-year retention — which typically happens as soon as the use case crosses into analyst-facing or customer-facing territory.

Enterprise Operating Workflow​

The licensed workflow surface below is included here so engineers can evaluate both the community baseline and the production operating model from one page.

AxonFlow Enterprise includes a dedicated SEBI module for audit-centric regulatory workflows. The current implementation is narrower than a full platform governance suite, but it is useful precisely where many securities and advisory teams need help most:

  • packaging audit evidence
  • checking retention posture
  • measuring readiness before an audit or review
  • exposing a compact compliance dashboard

Requires Enterprise license.

What The Current Module Covers​

The SEBI module registers five routes:

RouteMethodWhat it supports
/api/v1/sebi/audit/exportPOSTCreate a SEBI audit export
/api/v1/sebi/audit/export/{id}GETInspect export status and metadata
/api/v1/sebi/audit/retentionGETCheck retention posture by data type
/api/v1/sebi/audit/readinessGETReturn readiness score and checks
/api/v1/sebi/dashboardGETReturn a dashboard-style summary backed by readiness and module health

Identity handling is also important:

  • Tenant context (derived from Basic auth credentials) is the preferred tenant identifier
  • X-Org-ID is still accepted as a legacy fallback
  • X-User-ID is captured when present for export attribution and traceability

When Teams Use This Module​

The SEBI surface is especially useful for:

  • algorithmic trading or advisory environments that need audit evidence on demand
  • periodic internal reviews of retention and audit completeness
  • operating teams that need a quick readiness signal before regulator, internal audit, or board review
  • organizations scaling from basic logging to structured export and evidence workflows

Core Workflow​

The typical operating loop is:

  1. run governed traffic through AxonFlow so decisions, policy checks, and LLM activity are logged
  2. generate a SEBI export for a period, tenant, or incident window
  3. verify retention status for the relevant record classes
  4. review readiness findings and close gaps before an audit or inspection

Export Audit Data​

The export request shape is code-defined and supports date windows, output format, framework labeling, archive inclusion, PII redaction, and typed filters.

curl -X POST http://localhost:8080/api/v1/sebi/audit/export \
  -H "Content-Type: application/json" \
  -H "X-User-ID: compliance-ops-1" \
  -d '{
    "start_date": "2026-01-01T00:00:00Z",
    "end_date": "2026-03-31T23:59:59Z",
    "data_types": ["policy_violations", "llm_calls", "decision_chain"],
    "format": "xml",
    "framework": "SEBI_AI_ML",
    "include_archived": false,
    "redact_pii": true,
    "filters": {
      "agent_ids": ["advisor-agent"],
      "severity": "high",
      "include_model_info": true
    }
  }'

Why this matters for real teams:

  • xml is useful when the review process still expects structured regulator-facing artifacts
  • redact_pii helps when evidence must be shared beyond the narrowest internal reviewer group
  • typed data_types lets you package only the evidence set needed for an inspection, investigation, or quarterly review

Check Export Progress​

curl http://localhost:8080/api/v1/sebi/audit/export/exp_12345 \

The response includes an export_id, status, framework, summary metadata, and, for larger exports, a download_url plus expiry metadata when cloud storage is configured.

Check Retention Posture​

The retention endpoint is more useful than a simple "retention is enabled" flag. It returns per-data-type retention status, including configured days, oldest and newest record timestamps, total and archived record counts, storage usage estimates, and compliance status.

curl http://localhost:8080/api/v1/sebi/audit/retention \

This is the endpoint compliance and platform teams should use before claiming an audit window is fully defensible.

Check Readiness Before An Audit​

curl http://localhost:8080/api/v1/sebi/audit/readiness \

The readiness response is built around:

  • ready
  • score
  • checks
  • recommendations

Each check carries a name, description, status, and optional details, which makes this endpoint practical for an internal pre-audit checklist or change-management gate.

Dashboard For Platform Teams​

curl http://localhost:8080/api/v1/sebi/dashboard \

The dashboard endpoint is useful when platform or compliance operations want one compact view instead of stitching together retention and readiness calls separately.

For production teams running securities or advisory workloads, the strongest pattern is:

  • use policy enforcement and PII controls in the Agent
  • use the SEBI module for evidence packaging and posture checks
  • use HITL for high-value or high-risk review gates
  • use circuit breakers when an unsafe workflow or systemic issue needs immediate interruption

That keeps runtime governance and regulator-facing evidence connected instead of treating them as two separate projects.

What To Pair With The SEBI Module​

The SEBI APIs are strongest when paired with:

  • HITL & Circuit Breaker API Reference for human oversight and emergency shutdowns
  • enterprise deployment and monitoring docs so exports and retention checks are part of normal operations, not emergency-only workflows
Last updated on