Microsoft and 40 groups start an AI Transparency Initiative
Microsoft tells that an agent must not be the only proof of what it did, and that other systems must give proof too.
At Microsoft Ignite on Tuesday 17 November 2026, Microsoft CEO Satya Nadella showed the AI Transparency Initiative. It is work with customers and partners on one open method of security and control for AI models from all providers. More than 40 organizations are in it. The goal is that a company can see, check and control what its agents do.
How it works
The initiative has 3 parts:
- Microsoft joins the agent standards work of the US National Institute of Standards and Technology.
- It works with more than 40 organizations on open standards for AI that a third party can check.
- It adds the results to Microsoft AI, Foundry and Copilot, for use with all security tools.
Each record must show its source: the model, the software around it, or other systems such as identity and policy.
The proof
There are no test results and no standard at this time. Accenture, Adobe and KPMG gave quotes about the plan. Mastercard and Volvo Cars are also in it.
What is not known
Microsoft gave no date for the standards. It did not tell which parts will work with models from other providers.
Insight: Microsoft wants proof that does not come from the model
In October 2026, Nadella wanted tamper proof records of model work, and Microsoft planned AI agents that monitor other agents. At this time it tells that models cannot control themselves. Google and Cloudflare showed their own open agent checks that month. This looks like a race to set the record format that auditors accept. The format that wins decides what you must record.
Sources
Posted