AI/LLM Security Testing
Security assessment of LLM-powered applications, RAG pipelines, and AI agents mapped to the OWASP Top 10 for LLM Applications.
Core Focus Areas
The 6-Phase Workflow
1. Model Inventory & Threat Modelling
Cataloguing models, prompts, and tools to build a threat model mapping trust boundaries and the blast radius of LLM actions.
2. Prompt Injection Testing
Probing for direct and indirect prompt injection across user inputs to test system prompt leakage and persona escapes.
3. RAG & Retrieval Security
Assessing retrieval pipelines for data poisoning, embedding manipulation, and cross-tenant leakage.
4. Output Handling & PII
Testing downstream consumer handling of model outputs for XSS, code execution, and leakage of PII.
5. Guardrails & Abuse Review
Evaluating input filters and classifiers to test cost-amplification attacks and runaway agent loops.
6. Reporting & Recommendations
Delivering findings mapped to the OWASP LLM Top 10 with reproducible prompts and architectural mitigations.
What You Receive
Executive Summary
A comprehensive AI/LLM threat model and architecture review assessing structural risks.
Technical Findings
A findings report mapped to OWASP LLM Top 10, featuring reproducible prompt injection proofs and RAG security assessments.
Assurance & Verification
A targeted retest of all critical findings following the application of remediation steps.
Get Started
Pricing depends on your scope. After the scoping call, we review what you need and send you a clear quote for this service. NDA available.
Schedule Your Scoping Call