Leaks found
Find leaks of customer data and system instructions.
Adversarial testing of your chatbots and AI agents, before attackers find the gaps.
One instruction hidden in an email or web page can turn an AI agent against you. A traditional penetration test may not cover this.
Find leaks of customer data and system instructions.
Whether agents can be steered into harm, and limits if they can.
Show how AI risk is being managed, with test results to back it.
Who could attack it, how, and what is at stake.
Direct and indirect attacks through chat, documents, email and web content.
Exposure of personal information, secrets and system instructions.
What an agent can really reach and do on its own.
Poisoned content, cross-tenant access and permission bypass in search.
Prioritised fixes, help implementing them, and a retest to confirm.
Two to three weeks for one assistant or retrieval app, ending with a board summary and a ranked fix list.
from A$15,000
An attack where instructions hidden in user input, documents, web pages or emails steer an AI system into doing something it should not, such as revealing data or misusing a connected tool. Read our guide for leaders.
Yes. Models can memorise training records, and a membership inference attack can tell whether someone's record was used. In our research, differential privacy cut one such attack from 72% success to about 53%, close to guessing.
We agree scope, rules of engagement and test environments in writing first. Most testing runs against staging, and anything touching production is coordinated with your team.
Yes. Every finding comes with a practical fix, and we can implement the remediation and retest to confirm it worked.