AI application pentesting

Test the application around your AI.

Specialist delivery and availability are confirmed during scoping.

THE RIGHT STARTING POINT

For businesses launching or changing AI chat, enterprise search, copilots, and document workflows. Model internals and training pipelines require a separately scoped model or MLOps assessment.

A clear scope. A useful outcome.

A model response is one part of the application. Trace a request through authentication, tenant boundaries, retrieval-augmented generation (RAG), external tools, and output handling. Test realistic abuse cases with approved accounts and synthetic or authorized data, then give engineering owners reproducible evidence and a focused remediation plan. Agree application versions, testing windows, and retest scope before work begins.

The final scope, deliverables, and timing are agreed for your engagement.

What we cover

• Authentication, authorization, and tenant isolation
• Direct and indirect prompt injection across trusted and untrusted inputs
• RAG document permissions, retrieval leakage, and data exposure
• Output handling, integration permissions, and resource-abuse controls

What you take away

• Application threat model and agreed test coverage
• Validated findings with reproduction conditions and business impact
• Engineering remediation backlog with accountable owners
• Retest results, when included in the agreed scope

How to engage

An authorized application assessment followed by an optional remediation workshop and retest.

What progress looks like

Agreed abuse scenarios tested, access-boundary findings resolved, and critical fixes verified.

Before the engagement.

Agree the systems and business processes in scope, the authorisation required, the people involved and how the findings will be used. Timelines, deliverables, exclusions and any verification work should be confirmed directly.