Silverfort adds runtime identity control for AI agents built in Microsoft Copilot Studio

Silverfort, an identity security company, announced on June 8, 2026 that it has integrated its “Identity Security control for AI agents” with Microsoft Copilot Studio. The move targets a security gap that emerges when business teams use agent builders to create systems that do not just answer questions, but also take actions by calling enterprise tools and functions.

According to Silverfort, the integration evaluates agent access requests in real time and returns a decision before the requested action executes. In practical terms, when a Copilot agent attempts to use a configured connector or function, Silverfort’s controls are positioned as an inline runtime checkpoint rather than a post-event audit step.

Copilot Studio is Microsoft’s low-code environment for creating AI agents that can connect to company systems through predefined connectors. Microsoft documentation describes how Copilot Studio creates an “Agent ID” that defines the agent’s identity and is visible in an agent registry. Microsoft also explains that most connectors rely on on-behalf-of (OBO) authentication, meaning the agent acts using a signed-in user’s permissions, and that actions are intended to show up in audit logs with compliance context.

Microsoft further notes that, when a maker publishes an agent, the platform attaches API permissions to the agent’s Entra identity for the connectors the agent is configured to use, which administrators can govern through Microsoft Entra controls. In that context, Silverfort’s announcement frames its product as extending identity enforcement into the execution path of agent tool calls, with the goal of reducing the chance that an agent proceeds with an authorization decision that is too broad for the risk involved.

The integration lands as enterprises press for stronger controls as AI agents become more operational. Microsoft has separately described new capabilities for “real-time protection during agent runtime” in public preview, including integrations that let organizations connect monitoring systems or security platforms for runtime evaluation and control of agent behavior. Silverfort’s pitch is consistent with that direction, focusing on identity-based access decisions at the moment of action.

Still, details that matter for deployment are not fully described in the accessible announcement materials. Silverfort did not publicly specify which Copilot Studio connector categories are supported in the integration, how administrators configure identity policies within Silverfort, what performance or latency impact customers should expect, or whether the decisioning model aligns one-to-one with specific Microsoft Entra Conditional Access policies. The Microsoft portion of the rollout, including certification steps or any formal co-sell language, was also not included in the information retrievable from the original report.

For now, the key question is how organizations will combine Copilot Studio’s built-in authentication and governance model with Silverfort’s runtime enforcement. Buyers evaluating the integration may want to watch for follow-up technical documentation from both companies, including configuration guidance, audit-log behavior, and examples showing how tool access is allowed, blocked, or escalated under different identity and policy scenarios.