MICROSOFT has unveiled a bundle of measures to secure agentic AI and curb autonomous activity, including guardrails for AI models and new identity controls. At RSAC, it announced a preview feature in Microsoft Foundry to create guardrails and revealed agentic capabilities for its Security Copilot, while adding identities for agents in Entra ID to better track and govern non-human actors.
One key move is the launch of agent identities in Azure AI Foundry, giving AI agents a special Microsoft Entra ID identity managed by the company. Microsoft also broadened its guardrails with collections of controls for models or agents and expanded identity protections to shield data and resources from AI systems.
The company says you can monitor agents’ configurations and use an identity registry to flag risky capabilities, with Security Copilot and a new Security Analyst agent supporting deeper investigations using telemetry from Defender and Sentinel. According to analyst firm Omdia, more than half of companies surveyed lack confidence in securing resources accessed by non-human identities, underscoring the urgency of these guardrails.