One AI application, tested the way a real user would probe it. You get the evidence: what broke, what held, and a fix list. One fixed price, quoted on the scoping call.
You put a customer-facing chatbot or AI feature live. Now a customer’s security questionnaire, a bank’s third-party review, or a privacy assessment asks how it was tested and what it can reveal. A generic pentest report does not answer that question. The AI Red Team Check does, with the exact prompts and replies.
One chatbot or AI application, in one environment, under a signed test authorization.
Instructions that try to override what your AI was told to do.
Attempts to talk the AI past its rules and policy guardrails.
Whether customer records, internal notes or secrets can be pulled out.
Whether your hidden instructions can be exposed.
The way a patient user probes over a conversation, not a single prompt.
We plant a secret inside the system first, so a leak is proven, not guessed.
Steps to reproduce, the OWASP category and a severity for each one.
A coverage table with the exact prompts and replies, so a reviewer can see what was tried and what resisted.
The fixes in priority order for your engineers, and one page written for the person who signs.
The scoping call sets the goal and the end date. The readout walks through the findings. One retest within 90 days confirms the fixes.
This is a security test, not a certification. We are not an auditor and the report is not a compliance attestation. It is evidence you can hand to the person asking, and it can be filed in SecuritComply against the control a reviewer is checking.
Not included: tool or agent testing, a real retrieval pipeline, image or audio inputs, and classic web or API testing. If you need those, say so on the scoping call and we will tell you plainly who does them.
Prompt injection is number one on the OWASP Top 10 for LLM Applications 2026, the third year in a row (Help Net Security, Aug 6 2026). Among Canadian businesses using AI in the second quarter of 2026, 28.2 percent use virtual agents or chatbots (Statistics Canada).
Prefer email? [email protected]
· Or call 1 (647) 948-6769
· SecuritAI Technologies Ltd., Toronto.
It is a security test of the AI itself. A classic penetration test checks servers, networks and application code. This check sends your AI the prompts an attacker or a curious customer would, over real conversations, and records what it did. Keep your classic test; this one answers a different question.
A way to reach the AI application the way a user does, a test account if one is needed, and a signed test authorization. The scoping call settles the goal, the environment and the end date before anything starts.
One fixed price for one AI application, quoted on the scoping call once we know what we are testing. No subscription is required for the check.
You get the fix list. One retest within 90 days confirms the fixes held. If you want the same attacks blocked in front of the AI from then on, each finding can become a rule in the SecuritAI firewall; that is a separate decision, never a condition of the check.