Turn a finding into a test

Microsoft open-sourced RAMPART and Clarity on 20 May. RAMPART provides a framework for expressing adversarial and benign agent scenarios as repeatable tests that can run in continuous integration. Its stated aim is to keep a discovered failure represented in later regression checks.

Examine the intended behaviour too

Clarity takes a different role: it helps teams examine software-engineering assumptions before building. The announcement argues that agents with access to messages, records, code and external tools need a continuous safety process. These tools supply parts of that engineering workflow. Their release does not provide a certification for an agent, and a test set remains only as useful as the behaviours and boundaries it actually exercises.