Penetration Testing

AI & LLM Application Penetration Test

A chatbot or an agent built on a large language model (LLM) can be turned against you with words: a crafted message, or an instruction hidden in a document or web page it reads, can make it leak data or misuse the tools it is connected to. We test your AI application the way an attacker would.

REQUEST THIS SERVICE
Why choose us

We are an Italian security company focused on cloud and application security. Our services draw on years of penetration testing, consulting, training and research.

Solid methodology built by experts
International certifications
Fast-growing security company
MORE ABOUT US

FAQ

AI and LLM penetration testing: common questions

What clients usually ask us before we test an AI application.

Is this the same as AI-powered penetration testing?

No, they are two different services. Here we test your AI application: the model, its integrations and its data. With AI-powered penetration testing, we use an AI platform of our own to widen the coverage of a penetration test on your applications.

What do you need from us?

A test account for each user role is essential. Anything else is up to you and depends on the kind of test you want: with the system prompt, the models in use, the configuration or the documents behind the RAG, we can also examine the application from the inside; without them, we test it as an outside attacker would.

Can you test in production, and what about API costs?

A staging environment that mirrors production is the safest choice, but production works too, with agreed testing windows and no destructive tests. Either way, testing consumes calls and tokens on your account: before we start, tell us the limits we need to stay within.

How do you report an issue the model doesn't always reproduce?

We report it anyway. AI solutions are non-deterministic: the exact same prompt often won't trigger the same behavior every time, and the report says so clearly.

Which of the team's certifications relate to AI and LLM testing?

For AI and machine learning systems, the team holds the C-AI/MLPen (Certified AI/ML Pentester). The certifications that relate to the rest of the application, such as OSWE and GWAPT for web applications, are listed on our penetration testing page, and the full list is on our About page.