Microsoft is presenting a suite of security tools at the RSAC Conference in San Francisco this week designed to address what it describes as an emerging governance crisis in enterprise AI: autonomous agents that can act across systems, access sensitive data and make consequential decisions, often without adequate oversight.
The company's framing is direct. Vasu Jakkal, Microsoft's corporate vice president for security, wrote in the accompanying blog post that AI agents risk becoming "double agents," a phrase that captures the tension between the productivity benefits of autonomous AI systems and the novel attack surfaces and internal control failures they introduce.
The statistic Microsoft leads with is striking: 80% of Fortune 500 companies are already using AI agents, a pace of adoption that has significantly outrun the security and governance frameworks most organisations have in place to manage them.
The centrepiece of Microsoft's response is Agent 365, a control plane designed to give security and technology leaders visibility into agent behaviour at scale, covering what agents are doing, what data they are accessing and whether their actions fall within defined policy boundaries.
The broader suite integrates agent governance across Microsoft Defender for threat detection, Microsoft Entra for identity and access management, and Microsoft Purview for data governance, with specific features aimed at limiting data oversharing and securing the credentials agents use to operate across enterprise systems.
New capabilities include an AI security dashboard, network-layer protections against prompt injection attacks in which malicious instructions embedded in external content attempt to redirect agent behaviour, and data loss prevention controls embedded directly into Copilot workflows rather than applied as a separate layer.
Related reading
- Microsoft makes water utility cybersecurity programme permanent after pilot reveals sector's vulnerability
- Qurrent raises $15m to replace back-office workers with AI agents
- Nvidia launches NemoClaw to bring security infrastructure to always-on AI agents
Security Copilot is now included in Microsoft 365 E5 and E7 enterprise licensing tiers, and the company is adding automated triage and investigation agents inside both Defender and Purview to reduce the manual workload on security operations teams.
Microsoft said its security products currently help protect 1.6 million customers, one billion identities and 24 billion Copilot interactions, a scale it argues gives its threat detection systems an inherent data advantage over narrower alternatives.
The recap
- Microsoft unveils security stack for agentic artificial intelligence at RSAC
- Microsoft Security protects 1.6 million customers and 1 billion identities
- Agent 365 will enter general availability as announced
