Prompt layer
Can your agent be tricked
We test whether someone can get around your agent's rules and make it leak customer data or do something it shouldn't. It's the most common attack on AI agents today - prompt injection.
AI Trust Layer
for AI Agents
Security audits for AI agents and EU AI Act compliance. For companies already running AI, or planning their first rollout.
01 What we check
Three layers of security for your AI agent
Every agent you put in front of customers has three potential attack vectors. We audit all of them - from the prompt layer to the data layer.
Tool layer
What the agent can do
Every agent has permissions: it reads data, sends messages, connects to other systems. We check how far that access goes when something breaks, and make sure it can't do damage beyond what it's allowed to touch. The same agent-security principles Anthropic publishes.
Data layer
GDPR and the AI Act
We check how your agent handles your customers' personal data and whether it's GDPR-compliant. If it qualifies as a high-risk system, we add EU AI Act compliance - before someone else checks it for you.
02 Audit process
From first call to report in 5 days
We start with a 30-minute call, then take it from there. We test in a way that keeps your agent live for customers the whole time. You finish with a report, a fix list sorted by risk, and an hour to walk through it.
-
Day 1
Call and mapping
30 minutes online. We get to know your agent: what data it holds, what it can access, and where it meets your customers. This is the only step that takes your time.
-
Day 2-3
We try to break it
Automated attacks plus hand-written scenarios from your industry - the kind no off-the-shelf scanner would think of. This is where the agent's weak spots show up, and what they'd put at risk.
-
Day 4
AI Act compliance
We check your agent against the EU AI Act: what it already meets, what's missing, and what to fix. The scope depends on how risky the agent is.
-
Day 5
Report and session
A report: every gap, its risk, and a concrete fix, starting with the most urgent. Then an hour online - we walk through the report and you know exactly where to start.
Next move
Let's check your agent's security
Leave your details, we'll reply within 24h with a time for the first call.