Microsoft launches ASSERT open source tool creating scored AI tests
Microsoft just dropped ASSERT, a new open-source tool that helps developers check if their AI systems are actually behaving the way they're supposed to.
It takes simple descriptions, like "a document research AI agent shouldn't send emails to people outside the company" or "limit confidential information to C-level executives," and turns them into scored tests, making it easier to spot any issues before things go live.
ASSERT lets developers set custom rules
ASSERT is all about letting developers set custom rules and then automatically testing their AI for those behaviors. It tracks every step so failures are easy to catch.
Sarah Bird, chief product officer of Responsible AI at Microsoft, says ASSERT works across all stages, from building to monitoring, helping make AI more trustworthy.