Methodology — AI Platform VAPT — LLM & ML Penetration Testing

Guiding principles

• Risk-led, not checklist-led — every control traces back to a business risk.
• Evidence-first delivery — every workshop ends with an artefact in your repository.
• Local context — Arabic, English and Hindi delivery, local regulator relationships.
• Single source of truth — one risk register, one control set, many audits.

Phase 1. Threat Modelling

Map model inventory, data flows, agent tools and trust boundaries against OWASP LLM Top 10 and MITRE ATLAS tactics.

• Defined entry and exit criteria captured in the engagement charter
• Weekly progress reporting against an agreed traffic-light scorecard
• Outputs reviewed by a Lead Auditor before sign-off
• Lessons captured to refine the next Cybersecurity engagement

Phase 2. Adversarial Testing

Prompt injection (direct and indirect), jailbreaks, system-prompt leakage, training-data extraction, model DoS, RAG corpus poisoning and tool/function-call abuse.

• Internal audit dry-run with formal findings register
• Management review with executive sponsor
• External audit liaison and observation room support
• Findings closure plan with target dates and owners

Phase 3. Platform & Pipeline

Pentest the surrounding stack — vector DBs, inference endpoints, fine-tuning pipelines, MLOps tooling, model registries and identity boundaries.

• Defined entry and exit criteria captured in the engagement charter
• Weekly progress reporting against an agreed traffic-light scorecard
• Outputs reviewed by a Lead Auditor before sign-off
• Lessons captured to refine the next Cybersecurity engagement

Phase 4. Reporting

Risk-ranked report with PoCs, screen captures, repeatable payload library and CVSS-aligned scoring.

• Defined entry and exit criteria captured in the engagement charter
• Weekly progress reporting against an agreed traffic-light scorecard
• Outputs reviewed by a Lead Auditor before sign-off
• Lessons captured to refine the next Cybersecurity engagement

Phase 5. Re-test

Validation of remediated findings and tuning of guardrails, evaluators and detection rules.

• Internal audit dry-run with formal findings register
• Management review with executive sponsor
• External audit liaison and observation room support
• Findings closure plan with target dates and owners

Quality gates

Each phase ends with a formal gate review attended by the engagement partner, your sponsor and any second-line stakeholders. No phase closes until the gate criteria are documented and signed off.

• Gate 1 — scope, RACI and risk appetite formally agreed.
• Gate 2 — control design reviewed and approved by your security committee.
• Gate 3 — evidence pack independently sampled before audit submission.
• Gate 4 — post-audit lessons-learned and continuous improvement plan signed off.