AgenTrust: From Governance Decisions to Verifiable Agent Actions

Most agent governance tools can decide whether an action should be allowed, but they struggle to prove what was actually requested, authorized and executed across agents, tools and infrastructure. AgenTrust fills this gap by connecting declared capabilities, policy enforcement and tamper-resistant evidence into a shared trust layer, building on Microsoft’s Agent Governance Toolkit. This talk explores how agent manifests, governed tool boundaries and verifiable receipts can make autonomous systems accountable across organizational and technical boundaries. It also shows why this missing layer becomes essential as agents move from generating content to taking real-world actions.

About the Speaker
Imran Siddique is Chief Platform Officer at OPAQUE, where he leads the engineering organization and platform strategy for verifiable and confidential AI. Before joining OPAQUE in June 2026, he spent 18 years at Microsoft building and scaling platforms including SQL Azure, Azure DevOps, and Azure for Industries. He is the creator of the open-source Microsoft Agent Governance Toolkit and now leads work extending agent governance into cryptographically verifiable identity, hardware-enforced execution, and independently verifiable evidence through Agent Manifest, Confidential MCP, and the AgenTrust ecosystem. His focus is closing the gap between defining what AI agents are allowed to do and proving that those rules were actually enforced at runtime.