AI
AI security review
A security review of software that uses AI models: what the model can see, what it can do, and how someone could trick it.
What we check
Prompt injection
Whether text from users, emails, web pages or documents can override your instructions and make the model do something else.
Data exposure
Whether the model can see or reveal data the person using it shouldn't, including other customers' data.
What it can do
Tools and actions the model can call, and whether each one checks permissions on the server, not just in the prompt.
Logging and limits
Whether you can see what happened after the fact, and whether usage limits stop runaway costs.
How we report
We use the OWASP Top 10 for Large Language Model Applications as a checklist, test against your real system, and report each finding with how to fix it, most serious first.
Technologies
Questions
Can you review an app someone else built?
Yes. We need access to the code or a test environment, and we'll agree what we can and can't test before starting.
Tell us what you're building.
Email us about your project: what it needs to do, who it's for and when you'd like it live.