By Laura Hyde |
Microsoft has warned AI governance must evolve for the agentic era in its 2026 Responsible AI Transparency Report.
In the report Microsoft highlights how AI systems are becoming more capable of executing multi-step actions which is creating a new risk landscape in which factors such as access, permissions, memory and how AI is used over time are becoming increasingly important. Microsoft believes this shift requires a more dynamic approach to governance and has re-engineered its Responsible AI Standard.
The Responsible AI Standard has been redesigned to be “more adaptive” to the evolving AI technology stack and the role that Microsoft plays in developing or deploying AI products. It is now structured by reference to different components of the tech stack – models, platform services and applications – with “core requirements” that always apply, and more targeted, scenario-specific requirements capable of evolving as risk changes.
In the report, Microsoft also highlights the link between governance and engineering, with technical risk management practices now focused on systems capable of “retaining memory, using tools, accessing data and taking actions on behalf of users”. The report states “governance only works when people can put it into practice” which is why Microsoft has trained thousands of engineers and product managers on topics such as agentic AI threat modeling and prompt injection defenses.
The report reads: “Taken together, these efforts reflect a simple commitment: as AI becomes more powerful and more present in people’s lives, Microsoft will continue evolving governance, strengthening safeguards, and expanding the practical tools needed to earn and sustain trust.”
The company has also introduced a new Deployer Chapter covering the use of third-party AI applications within Microsoft. It requires teams to assess an application’s capabilities, limitations and safeguards; review deployment-specific risks; monitor performance and incidents after launch; and inform users about intended uses, limitations and safeguards.
“Responsible AI is not just about anticipating known risks; it is about building governance that remains effective as the technology, its uses and societal expectations continue to evolve,” said Natasha Crampton, chief responsible AI officer at Microsoft in the report’s foreword. “Trust is essential to unlocking AI’s benefits, and we remain focused on earning and sustaining it through effective governance – disciplined action, continuous improvement, and openness about our progress and priorities.”
Read the full 2026 Responsible AI Transparency Report.