AI assurance review
Before an assistant or agent reaches staff or customers, we test agreed misuse scenarios. We examine the application, model instructions, connected tools, data access, and retrieved documents, then report demonstrated behaviour and proposed changes.
Timeline agreed with scope. Fixed price agreed before work starts.
What you receive
- An attack-surface map covering the application, model instructions, connected tools, and retrieved documents
- Demonstrated findings with steps to reproduce them
- Findings mapped to a published industry reference list
- Prioritized proposed fixes and a re-test of agreed changes you apply
Scope and limits
We agree authorization, the system, environment, testing window, and re-test scope in advance. Testing uses a non-production environment unless you authorize otherwise in writing. Applying fixes, operating the system, and ongoing monitoring are separate work. Findings describe what was tested during the agreed window, not the system's overall safety.
When this is not the right engagement
If you need continuous monitoring, a review of every possible model output, or a formal audit opinion, this bounded testing engagement is not a substitute. Testing also needs an authorized environment and a system owner able to respond to findings.