Microsoft wants companies to stop relying on AI agents to explain their own behavior. The company says businesses need separate records that can check what an agent was asked to do and what it actually did.
Chairman and CEO Satya Nadella announced the AI Transparency Initiative at Microsoft Ignite on Tuesday, Nov. 17, 2026. The company described it as the next phase of its work on making AI safer.
The core idea is simple. Microsoft says that as AI systems act with more freedom, they need to be more transparent.
The Gap Microsoft Wants to Close
Many companies now connect AI models to data, tools, passwords and computer systems. This lets agents complete tasks without a person approving every step.
According to a Microsoft press release, current security tools can watch much of the activity around a model but cannot see what happens inside it. Companies also cannot independently confirm a model’s account of its own actions.
Microsoft says this creates risk for security, compliance and audits. It compares agents with access to data and credentials to a trusted insider who works at machine speed and scale.
The company also says responsibility stays with the business. When something goes wrong, leaders still answer to customers, workers, boards, auditors and regulators.
“For security controls to be effective, every organization needs greater transparency into the model itself,” said Hayete Gallot, Executive Vice President of Microsoft Security. She said the work aims to make containment stronger and improve traceability.
How the Initiative Will Work
The plan has three parts. First, Microsoft will expand its role in an agent standards effort run by the National Institute of Standards and Technology and use that research to help build open industry standards.
Second, Microsoft is working with more than 40 organizations, including Accenture, Adobe, KPMG, Mastercard and Volvo Cars. The group aims to develop standards for AI systems that can be checked independently, no matter which model or provider is used.
Third, Microsoft will add this work to Microsoft AI, Microsoft Foundry and Microsoft Copilot. Customers will be able to use the resulting records with their own security and compliance tools, including tools from other vendors.
The framework is built around three goals: see, verify and control. Companies should know what an AI was asked to do, which model it used, what data it touched and what actions it took.
Under the verify goal, records should show where each piece of evidence came from. That way, what a model reports can be compared with records from separate systems such as identity and policy tools.
The control goal focuses on letting companies enforce rules before major actions happen. Microsoft says this evidence should stay with the customer even if it switches models or providers.
The approach is meant to cover both closed and open-weight models. This would let developers update the models behind their agents without weakening oversight.
Partners backed the effort in their own statements. Aanchal Gupta, Chief Security Officer at Adobe, said security teams need clear visibility into what agents access and do.
John Israel, Global Chief Information Security Officer at KPMG International, said clients should not have to choose between innovation and safety. Harpreet Sidhu, global cybersecurity lead at Accenture, said an open, industry-driven approach can help companies secure AI models.
Microsoft says organizations that want to help shape the initiative can sign up through its website. More updates are expected during the Ignite conference.