Agentic Testing

Testing agentic systems is significantly more onerous due to the need to test different components within the system. We developed the Agentic Risk & Capability Framework help organisations and developers systematically identify and manage the safety and security risks of agentic AI systems.

Agentic Risk & Capability Framework

The Agentic Risk and Capability Framework defines (i) baseline risks (i.e., risks that apply to all agentic systems), and (ii) a hierarchical taxonomy of capability risks (i.e., risks related to specific capabilities that particular systems may have). Each risk can then be systematically tested and evaluated. The framework also provides a mapping of each risk to a set of technical controls, as well as an example implementation of the framework.

How to perform testing?

Work in progress!

We are currently in the process of developing a testing framework. Please check back soon!