Sample AI report · 35 pages · PDF
What an AI penetration test actually finds.
Plenty of firms will sell you an AI security checklist. Almost nobody will show you the deliverable. This is the whole thing: nine findings written out in full, across the administration plane, tenant isolation and the agent tool boundary — and not one of them is a jailbreak string.
What’s inside
Nine findings, and where they actually were
Executive summary
Where the risk actually sat — and why the model itself was not the problem.
Scope and methodology
The six areas an AI assessment has to cover, mapped to the findings they produced.
The administration plane
Who can write the prompt configuration, the pipeline credentials and the model lifecycle.
Tenant isolation
Whether one tenant’s users can reach another’s conversations, documents and embeddings.
The agent tool boundary
What an agent can be steered into calling — scored on what the tool reached, not on the injection working.
Remediation plan
Why the answer was one enforcement decision rather than nine separate patches.
Why people ask for it
Because the market is all checklists and no deliverables
Search for AI penetration testing and you will find a great many OWASP LLM Top 10 checklists and very few examples of what anyone actually hands over. This is the deliverable, so you can judge the work rather than the marketing.
It is a composite sample: assembled from the shape of several anonymised engagements, with synthetic illustrative data throughout. It describes no single real client, and no real host, tenant, person or credential appears in it. No vendor or platform is named, deliberately.
The ordering is the argument. Prompt injection is in the report, but it is rated on what it reached — not on the fact that it worked.
- Scoping an AI assessmentUse the six coverage areas as the basis for your own scope.
- Judging a proposalCompare it against what a prompt-injection-only test would have reported.
- Briefing engineersShow the team what an AI finding looks like with evidence attached.
- Briefing leadershipThe executive summary is written for people who sign things off.
Want this for your own AI features?
Tell us what you want to protect. A senior tester will help you agree the scope and next steps. We usually reply within one business day.
Confidential from the first conversation
We can sign your NDA before you share sensitive details. Your enquiry is covered by our privacy policy.
How we protect your information · Privacy policy