Microsoft Unveils Robust Security Framework for AI Agents at RSAC, Enhancing Enterprise Protection
March 22, 2026
AIP v0.5.52 has undergone extensive testing with hundreds of tests and agents, while ongoing W3C DID method registration is pursued, with links to GitHub, PyPI, and Trust Observatory for further exploration.
Security capabilities extend to endpoints, cloud workloads, and AI services, including Entra Internet Access prompt injection protection (GA by late March), Defender for Cloud updates with stronger container security and broader AWS/GCP coverage, and Defender predictive shielding for adaptive access during active attacks.
Shadow AI Detection was added via Entra Internet Access, enabling network-layer identification of unknown AI applications with a GA date set for March 31, 2026.
Unified Identity Security seeks end-to-end coverage across identity infrastructure, control plane, and threat detection/response in a single dashboard.
The company will showcase demonstrations at RSAC to present these tools as a cohesive security stack, emphasizing protection for AI agents and data across the enterprise.
Entra enhancements advance hardened identity infrastructure and intelligent access decisions, featuring backup/recovery, tenant governance, passkey improvements, external MFA, adaptive risk remediation, and a unified identity security dashboard with ITDR-focused Defender insights.
Microsoft has unveiled a comprehensive security strategy for agentic AI, spanning governance, identity protection, data security, and cloud and endpoint defense, designed to shield AI agents and data across enterprise environments.
Entra capabilities also include Entra Backup and Recovery (preview), Entra Tenant Governance (preview) to surface unmanaged tenants, updated passkey features, external MFA (GA), adaptive risk remediation (GA in April), and a unified identity security dashboard with an identity risk score in preview.
New Entra capabilities aim to strengthen resilience and governance around identities, and Purview features will prevent sensitive information from being processed by AI systems.
Additional feature timelines include Entra adaptive risk remediation becoming generally available in April, Enhanced Intune App Inventory due in May, and Purview embedded in the Copilot Control System planned for April.
A notable gap remains: cross-platform and cross-organizational agent identities are not yet fully addressed, including agent-to-agent trust, portable identities, and decentralized verification.
Agent Identity Protocol (AIP) is proposed as a cross-platform, DID-based solution to enable interoperable, cryptographic identities for agents across clouds, using Ed25519 keys, DIDs, and a trust graph.
Summary based on 4 sources
Get a daily email with more Tech stories
Sources

Microsoft Security Blog • Mar 20, 2026
Secure agentic AI end-to-end
DEV Community • Mar 22, 2026
Microsoft Just Launched Agent 365 and Zero Trust for AI at RSAC 2026. Identity Is Still the Foundation.
