Microsoft has begun an AI Transparency Initiative, a joint effort with customers and partners to set an open, shared approach to securing and governing AI across different models and providers. According to Microsoft, Chairman and CEO Satya Nadella announced it on Tuesday, Nov. 17, 2026, at the Ignite event.

Advertisement

The gap Microsoft describes

Microsoft says organizations can see much of the activity surrounding AI models but frequently cannot piece together what occurred between a request and its result, which makes independent verification of model behavior difficult. Enterprises are linking models to data, tools, credentials and infrastructure, so agents can act without a person approving each step. In Microsoft's account, current controls cannot see inside the models or confirm what the models report about themselves.

The company calls this an evidence gap that produces security, compliance, audit and accountability risk. It says agents holding such access resemble a privileged insider in risk terms, yet work at machine speed and scale.

Three parts of the plan

  • Microsoft will take a bigger role in the SI Agent Standards Initiative run by the National Institute of Standards and Technology, and use its research toward open, industrywide standards for AI transparency, security and governance.
  • Microsoft says it is collaborating with more than 40 organizations, among them Accenture, Adobe, KPMG, Mastercard and Volvo Cars, on open standards intended to make AI systems accountable and independently verifiable, whatever the model or provider.
  • The work will be folded into Microsoft AI, Microsoft Foundry and Microsoft Copilot, letting customers pair the resulting evidence with security and compliance tools from Microsoft or elsewhere.

Microsoft says the design is meant to span closed and open-weight models, so development teams can change the models behind their agents while IT, security and compliance staff keep their ability to govern and audit them.

See, verify, control

Microsoft frames its stance as three principles. Organizations should see what AI systems access and do, check those actions independently, and control what the systems may do. On verification, Microsoft argues that agents cannot be the only source of truth about their own behavior. On control, it says the evidence should remain with an organization if models or providers change.

Hayete Gallot, Executive Vice President of Microsoft Security, said the work will bring more transparency to all models and help inform accountability and traceability. The announcement also carries statements of support from Aanchal Gupta of Adobe, John Israel of KPMG International and Harpreet Sidhu of Accenture. The material is a company announcement, and the standards it describes are still to be developed.

Advertisement