Agent as a service
Prompt Injection Audit
We try to break your agent, then tell you how we did it
by Raghav Menon
What it does
A person plus tooling, run against your deployed agent. We attempt indirect prompt injection through every surface it reads — documents, web pages, tool output, user input — and report what got through, with the exact payload.
Delivered as a written report and a working proof for each finding. Retest included once you have shipped fixes.
Ideal use cases
- Agents with tool access that read user-supplied or web content
- Pre-launch review before an agent touches customer data
Where not to use it
- Closed systems reading only trusted internal input
- Teams with no capacity to act on the findings
The problem it solves
Agents that read untrusted content are the new attack surface, and most teams ship one without ever having been attacked.
Details
- Domain
- Category
- Sub-category
- Architecture
- Built for
- Runs on
- Protocols
- Ecosystems
- Deployment