Service
AI agent tool and permissions review
Most agent incidents come down to an agent being able to do more than the task needed. We map what your agent can reach, then test what it does when a tool fails or a task cannot be completed as asked.
What we test
- Tools, connectors and MCP servers the agent can call
- Workarounds when a tool is blocked or a page will not load
- Actions taken without being asked
- Confirmation steps in front of irreversible actions
What you get
A permissions map, a log of every unrequested or out-of-scope action observed, and a least-privilege configuration to replace the current one.
Where the fix is in the setup (a prompt, a permission, a retrieval or guardrail setting) we make it and run the failed cases again. The test suite is yours to keep.
Who asks for this
Engineering leads deploying coding, browsing or operations agents with real credentials.
Contact
Tell us what your assistant does. We'll tell you how we'd test it.
Other work
- Prompt injection testing for AI agents
- Third-party LLM red teaming
- Independent AI assistant assessment report
- Chatbot hallucination and accuracy audit
- EU AI Act transparency testing for chatbots
- Custom evaluation datasets and grading rubrics
- System prompt and data leakage testing
- Crisis and vulnerable-user response testing
- AI vendor due diligence testing
- Regression testing for model upgrades