AI you can trust and defend
We review and protect the AI tools your business uses or builds, from data access to prompt injection, and help you set clear rules for how your team uses AI.
Review your AI toolsWhat this service is about
What it is
A review and protection of the AI tools your business uses or builds, so they do not leak data or take unsafe actions. We also write clear rules for how your team uses AI and help you meet the obligations that apply to you.
Who it's for
Your team already uses ChatGPT or other AI tools, and you do not know with what data.
You are launching an assistant or AI feature and want it tested before customers use it.
A client, partner, or regulation asks how you manage AI risks.
What you get
An inventory of the AI tools in use and the data they touch
A security test based on the OWASP Top 10 for LLM applications
An AI usage policy in plain language for your team
A report with the changes made and what to watch next
What it looks like in practice
Illustrative examples of typical projects.
- 01
A chatbot tested before launch
A company is about to publish a customer chatbot. The review tries to make it reveal internal data, ignore its rules, or take actions it should not, and the issues are fixed before launch.
- 02
Unapproved AI tools under control
Employees use several AI tools on their own. The inventory finds them, the policy sets which ones are approved, and sensitive data stops going to personal accounts.
- 03
Ready for a client's questionnaire
An enterprise client asks a software company how it handles AI and data. A clear policy and documented controls answer the questionnaire.
Our Process
Inventory
The AI tools in use, official or not, and the data they touch.
Test
Each tool checked for leaks, misuse, and unsafe behavior.
Fix
Controls, guardrails, and policy, sized for a small team.
Hand off
A report, the changes made, and what to watch next.
Timeline and pricing
- Typical timeline
- A review of one AI tool usually takes one to three weeks. A usage policy and the basic controls for a small team usually take two to four weeks.
- How it is priced
- A fixed price based on the number of tools reviewed and the scope of the policy work. Periodic reviews can be agreed afterwards, as your tools and the rules change.
Common questions
An attack where someone hides instructions in a message, document, or web page to make an AI tool ignore its rules, leak data, or take unwanted actions. It is the top risk on the OWASP list for AI applications.
No. We help you understand which obligations apply to you and put the controls in place. Formal certification and legal advice come from the appropriate bodies and lawyers.
Often, yes. The main risk is what your team pastes into it. A short policy and the right account settings prevent most problems.
Only what the review requires, agreed in writing beforehand. We work in test environments whenever possible.