Skip to main content
SEBI May 2026 Advisory · CERT-In Empanelled Since 2008

AI-Resilient VAPT
for the AI threat era.

AI tools like Claude Mythos can find and exploit vulnerabilities at speed and scale. SEBI's May 2026 advisory says regulated entities must adapt. B-52 — our AI-powered pentesting and red-teaming platform — has been doing exactly that since long before SEBI named it.

10/10
Annexure-A directives addressed
6,700+
B-52 powered engagements
370+
BFSI engagements
Since 2008
CERT-In empanelled

Why this service exists

SEBI just named the threat. We've been hunting it.

The SEBI May 2026 advisory (HO/13/19/12(1)2026-ITD-1_CIMGI/10873/2026) cites Claude Mythos as a representative AI tool that "may give rise to heightened risk exposure by enabling identification and potential exploitation of existing vulnerabilities using speed and scale." The regulator wants regulated entities to recalibrate their cyber posture for an AI-augmented attacker. We help you do exactly that.

Track 1

AI-augmented attacker scenarios

Your stack tested against the threat profile SEBI just flagged.

  • AI-driven recon at speed

    Mythos-class tools enumerate, fingerprint, and dependency-map a target stack in minutes. We test whether your perimeter and detection capacity hold up against this telemetry pace.

  • Agentic exploit chaining

    Multi-step attack chains assembled by AI agents — credential stuffing into IDOR into privilege escalation. B-52 maps 5–15 such chains per typical engagement and validates each one is exploitable.

  • AI-augmented business-logic abuse

    Pattern-matching across application flows surfaces subtle authorisation, transaction, and workflow bypasses that scanners and junior testers miss.

  • Mass parameter and endpoint discovery

    AI tooling can spider, fuzz, and parametrise faster than legacy scanners. Our discovery phase runs on the same class of capability — so your blind spots surface before an attacker finds them.

Track 2

AI-system-defender scenarios

Your AI-ML systems tested as defenders against abuse, manipulation, and integrity attacks.

  • Prompt injection across customer-facing AI

    Chatbots, RAG-powered support, agentic workflows, and embedded LLM features tested for instruction smuggling, system-prompt extraction, and indirect prompt injection.

  • Model and training-data integrity

    Tested against output manipulation, model abuse, training-data leakage, and adversarial inputs that compromise the "reliability of outputs" the SEBI advisory specifically calls out.

  • AI-ML pipeline supply chain

    Inference endpoints, vector stores, embedding caches, and model artefact stores reviewed against the same threat model as production application stacks.

  • Agentic workflow safeguards

    Permission boundaries, action-filter enforcement, and audit-trail integrity tested for AI workflows that take real-world actions on customer data.

Engagement shape

Flexible across RE sizes — including smaller entities.

The advisory addresses 19 RE categories. Some are MIIs and large AMCs with dedicated SOC functions and locked production change windows. Others are KRAs, IAs, RAs, smaller depository participants, and merchant bankers running on lean tech teams. The standard 12-week engagement isn't right for everyone.

We can phase the work — Phase 1 gap assessment as a fixed two-week scope, Phase 2 VAPT separately — so smaller REs can pace the spend without losing audit-trail coverage. Tell us your RE category and your customer-facing surface in the scoping call; we'll propose a shape that fits.

For MIIs and entities with locked change windows: the Phase 1 gap assessment is read-only and produces an evidence pack the IT committee can review without any production change. Phase 2 testing is scoped to your change windows by design — no surprise deploys, no after-hours request that hits a market-day blackout.

Common engagement scopes

Engagements shaped to your RE category

AI-Resilient VAPT engagements cluster into well-defined patterns — sized for the SEBI advisory's 19 RE categories and your actual attack surface.

Small RE single-application KRA, IA, RA, small DP/merchant banker with one customer-facing app. 4-week focused cycle.
Mid-sized RE multi-system AMC, broker with web + mobile + API. 8-week VAPT cycle with optional Phase-1 gap assessment.
Large RE full-stack Bank, large AMC, depository with web, mobile, API, backend, and internal systems. 12-week full cycle.
MII with locked change windows Read-only Phase 1 + testing scoped to production change windows. No surprise deploys, no market-day blackouts.
AI-system-only assessment Focused on AI-ML pipeline, agentic workflows, and LLM surfaces. Venue for entities with existing WAPT coverage.

Phase flexibility

Pace the spend without losing audit-trail coverage

Smaller REs can separate the two-week Phase 1 gap assessment from the VAPT itself — schedule them months apart if needed. Audit trail remains complete. Larger REs run both phases back-to-back with SOC/SOAR integration support in weeks 7–12.

Phase 1 only

Gap assessment + evidence pack. IT committee reviews before committing to Phase 2.

Phase 1 + Phase 2 (4 weeks)

Gap assessment + focused VAPT. For entities with mature CSCRF coverage.

Full 12-week cycle

Gap assessment → VAPT → SOC/SOAR support → IT-committee package.

Methodology

B-52 inside. Senior auditors on top.

Three phases. Discovery, testing, delivery. Same rigour every engagement.

Discovery

01

Scope + AI threat-model intake

Map your stack, customer-facing AI surfaces, third-party COTS dependencies, and SOC log sources. Calibrate the AI threat model to your business.

02

B-52 phase-1 fingerprinting + SBOM

Per-application fingerprint, endpoint enumeration, parameter discovery, and SBOM with CVE applicability mapping.

03

Coverage validation

B-52 cross-references discovery artefacts (mind maps, spider, JS analysis, route files, server logs) to surface missed attack surface before testing begins.

Testing

04

AI-augmented attacker simulation

Speed-and-scale recon, exploit chaining, agentic abuse — tested against your stack with B-52 generating 2,000+ test cases per engagement, scoped to your application surface.

05

AI-system-defender testing

Prompt injection, model abuse, training-data leakage, output integrity tests across your customer-facing AI surfaces and AI-ML pipelines.

06

Attack-chain analysis

5–15 multi-stage attack chains mapped per typical engagement. Privilege escalation, lateral movement, and business-impact quantification.

Delivery

07

Three-layer expert review

L1 Auditor → L2 Senior Consultant → L3 Security Architect. Every finding verified exploitable before delivery.

08

Annexure-A coverage map + IT-committee package

Executive summary, technical report, Annexure-A coverage matrix, M-SOC readiness brief, and the long-term AI defence plan that satisfies item 10.

Coverage matrix

All 10 Annexure-A directives. Mapped to deliverables.

Each SEBI directive maps to a specific Security Brigade deliverable. Items 6 (M-SOC onboarding) and 10 (long-term AI plan) include support and input-package framing — the RE retains ownership of those endpoints.

# Advisory directive Security Brigade deliverable
01 Patch management + virtual patching Prioritised patch matrix + WAF/ACL recommendations
02 AI-based + conventional VA Full B-52 powered VAPT with verified-exploitable findings only
03 Third-party + COTS vendor risk Vendor risk assessment + ShadowMap continuous monitoring
04 Change-management testing Change-driven retests + impact analysis built into Lemon delivery
05 API security (inventory, authN/Z, rate limit, whitelist) API inventory + OWASP API Top 10 coverage
06 SOC monitoring + SOAR/SIEM + M-SOC onboarding M-SOC onboarding readiness support (M-SOC onboarding itself runs between RE and NSE/BSE), SOAR playbook design, red-team validation
07 CSCRF risk assessment with AI scenarios AI-threat-model addendum to CSCRF risk register
08 ZTNA + system hardening CIS-benchmark validation + identity-perimeter testing
09 Asset inventory + SBOM Per-application SBOM with CVE applicability mapping (every engagement)
10 Long-term plan for AI in detection + agentic mitigation Input package for the IT committee (the committee owns ratification)

Deliverables

What you get

Reports for three audiences — the IT committee that needs the risk picture and AI defence roadmap, the CISO/SOC team that needs technical findings and M-SOC readiness brief, and the SEBI auditor who needs the Annexure-A coverage matrix.

Executive Report

Risk overview, critical findings by business unit, Annexure-A coverage summary. IT-committee-ready.

Technical Report

Step-by-step POCs, screenshots, request/response data, CVSS, attack-chain maps, and code-level fix examples.

Annexure-A Coverage Matrix

Per-directive mapping of findings to SEBI advisory items 1–10. CSCRF cross-reference for each finding.

Security Certificate

Formal certificate for compliance, customer assurance, SEBI auditor submission, and vendor due diligence.

FAQs

Common questions from CISOs and IT committees.

How is AI-Resilient VAPT different from a regular pentest?

A regular pentest assumes a human attacker. AI-Resilient VAPT assumes an AI-augmented attacker (Mythos-class) operating at speed and scale, AND tests your AI systems as defenders — prompt injection, model abuse, agentic-workflow safeguards. The methodology is the same B-52 6-phase audit, recalibrated for the AI threat surface SEBI flagged in May 2026.

Does this engagement satisfy the SEBI May 2026 advisory?

It addresses all 10 Annexure-A directives. Items 1, 2, 5, 7, 8, and 9 are delivered through the VAPT itself. Items 3 (third-party risk) and 6 (M-SOC + SOAR) are delivered through paired services and readiness support — the actual M-SOC onboarding is between the RE and NSE/BSE. Item 10 (long-term AI plan) is delivered as an input package for the IT committee, which retains ownership of the plan. We also map every finding back to the CSCRF baseline.

Do you need direct access to our LLMs and AI-ML pipelines?

Yes for AI-system-defender testing. We test prompt injection and model behaviour against your actual deployed AI surfaces — chatbots, RAG endpoints, agentic workflows, vector stores, and inference endpoints. We work with your team to scope authorised access and provide a clean test plan before engagement start.

How long does an AI-Resilient VAPT engagement take?

A typical 90-day cycle: weeks 1–2 gap assessment, weeks 3–6 full B-52 powered VAPT, weeks 7–9 SOC/SOAR/M-SOC integration support, weeks 10–12 long-term AI defence plan and IT-committee package. Scope flex is available — a focused 4-week engagement is possible for entities already CSCRF-mature.

Is this just for SEBI-regulated entities?

No. Banks under RBI cybersecurity guidelines, insurers under IRDAI, healthcare under HIPAA, payment processors under PCI DSS — all face the same AI-augmented attacker threat. The SEBI advisory is the most explicit regulatory framing in India to date, but the underlying threat applies cross-sector.

What is B-52 and is it included?

B-52 is Security Brigade's AI-powered pentesting and red-teaming platform. It runs on every engagement — generating structured test plans, validating coverage, mapping attack chains, and verifying exploitability before findings reach the report. It is included in the engagement, not a separate purchase.

Scope your AI-Resilient VAPT engagement.

Phase-1 gap assessment is available as a fixed two-week scope, separate from the full 12-week cycle. Tell us your stack and your timeline; we propose a structure that fits both.

Request a Scoping Call