Inputs and retrieval
- Direct and indirect prompt injection
- System prompt and instruction exposure
- Retrieval and memory poisoning
- Model and application supply-chain risks

AI security assessments
Security assessments for AI applications, copilots, and agents. See where the system could expose data, exceed permissions, or take actions you didn’t intend.
We agree on which parts of your AI application are in scope, including the model, its data sources, and the systems it can affect.
We can build a custom evaluation harness for your AI application so the same security and safety checks can be repeated as models, data sources, and features change. Choose a monthly or quarterly review cycle and agree up front on the standards and requirements you want to assess.
Request a recurring assessmentIdentify users, data, tools and approval requirements before testing.
Test agreed scenarios and verify business impact within the engagement boundaries.
Review a formal report and walk through the changes that reduce risk first.
Tell us what it does, what it can access, and when you need the assessment. For public-facing infrastructure, explore our external tests from multiple attacker perspectives.