Building Agents You Can Trust on Windows | BRK262
Windows is adding platform-layer controls to make AI agents act safely and predictably, not just capably.
“we cannot rely on the agent to self-govern”
At Microsoft Build (BRK262), the Windows platform team argued that agentic AI's growing autonomy creates a large 'blast radius' when agents act wrongly (deleting databases, ignoring stop commands), and that prompting alone cannot fix this. Microsoft is positioning Windows to provide OS-level guardrails so agents run safely and predictably, signaling that agent trust and containment is moving into the platform layer.