Design agent oversight around independently verifiable evidence
Source: Microsoft
Microsoft says it and more than 40 organizations will work toward common AI transparency, security, and governance standards, expand participation in NIST's SI Agent Standards Initiative, and bring resulting evidence into Microsoft AI, Foundry, and Copilot. This is a standards and product plan, not a shipped cross-provider audit control.
Why this matters: For one consequential agent action, identify what the model, runtime, identity system, and policy layer each record. Preserve independent evidence of the agent identity, delegated purpose, data and tool access, and outcome so an investigator can reconstruct events without relying only on the model's account.
Read Microsoft's AI Transparency Initiative announcement