We keep attacking your AI agent, so attackers find less.

The problem

An AI agent changes every week: new prompts, new tools, new models. A test you ran once goes stale fast.

How it works

1. 01

   You set the scope

   You say which AI agent we may attack, and what is off limits.

2. 02

   We attack inside it

   Our team attacks the agent inside that scope.

3. 03

   You see what got through

   You see which attacks worked.

4. 04

   It feeds your defence

   What got through goes into your defence.

What we attack

• Jailbreaks.
• Prompt injection through tools and documents.
• Data leaks.
• Dangerous actions, the area Auto Mode covers.

Chaos Mode tests AI systems only.

Works with EU Mode

Chaos Mode is part of EU Mode. What gets through feeds your defence: HaloGuard EU and Auto Mode.