When should agent security testing happen?
Before production access, when a new tool or integration is added, or when an agent’s permissions expand. Changes to models, instructions and oversight can also justify a new assessment.
Services /
AI Agent Security Testing
Test what your agent can read, change and trigger, and whether its permissions match the job it has been given.
Discuss your testing needsWhat the work covers.
For organisations connecting AI agents to enterprise tools and vendors shipping products that can take action on a customer’s behalf.
An agent that can send a message, update a record or call a tool creates a different risk from a system that only produces text. Witcher examines the path from an instruction to an action: which identity is used, what access is available, how untrusted content influences the decision and where a person is expected to intervene.
- 01Tool permissions and privilege boundariesCheck whether tool access is limited to the intended task, user and resources, including attempts to cross roles or obtain broader access.
- 02Untrusted content reaching actionsTest whether instructions in documents, tool responses or other inputs can cause the agent to take an unintended action.
- 03Human approval and escalationExamine when the agent asks for approval, what the reviewer can see and whether the approved action matches what is executed.
- 04Failures across a workflowInvestigate how ambiguous inputs, incomplete tool responses and chained actions affect the agent’s behaviour and ability to stop.
Evidence your team
can act on.
A map of the tested action and permission boundaries, technical findings with supporting evidence, and recommendations for access controls, oversight and remediation.
Scope, assessment criteria, test access and deliverables are agreed before work begins. See our assurance methodology.
Questions before
you start.
Can this cover an agent already in use?
Yes. We agree a test environment and boundaries appropriate to the system. Continuous assurance can provide a cadence for reassessing changes after the initial engagement.
Part of a wider
assurance engagement.
Explore how this work fits into Continuous AI Assurance, or compare all assurance services.
Testing for your
AI system.
Explore specific testing needs within an assurance engagement.
01
AI Security Testing
Understand how your AI application handles hostile inputs, sensitive data and the access it has been given.
02LLM Red Teaming
Challenge your LLM application with adversarial scenarios before those failures affect users or customers.
03AI Agent Security Testing
Test what your agent can read, change and trigger, and whether its permissions match the job it has been given.
Scope your AI testing.
Tell us about the system, the access it has and the decision the evidence needs to support.