Thursday, November 19, 2026, 00:24
Home»AI News»Microsoft Perception puts guardrails around security agents...
RSS

Microsoft Perception puts guardrails around security agents

Microsoft Perception puts guardrails around security agents

Perception brings together security signals, information about a customer’s environment and agents that can investigate or respond. Its controls can contain devices, revoke identity access and block agents at the authorisation layer. Microsoft says its own and partner sources provide more than 165 trillion signals a day through more than 500 connectors. Threat intelligence enriches the data before a context layer prepares it for agents. Microsoft says that layer is optimised for precision, recall and token efficiency; those are vendor claims, not independent performance findings.

Teams set incident criteria and limits for automated playbooks. Under Microsoft’s stated model, agents can investigate and act within those boundaries, but high-impact actions require human sign-off. MCP and API connections can extend orchestration to existing security and IT service management systems. Signals from AWS, Okta and Salesforce are available in preview.

Microsoft Defender’s ISOC combines security information and event management with behavioural analytics, threat intelligence and response capabilities. Its generally available Alert Triage Agent and Attack Investigation Agent are designed to handle investigative work. Microsoft says they can reduce time spent detecting and responding to threats, allowing analysts to focus on decisions requiring judgement.

Other tools have not reached general availability. The Web App Pen Test Agent and Codename MDASH, Microsoft’s code scanner, are both in preview. MDASH scans can feed the testing agent to help prioritise and fix vulnerabilities; MDASH also runs in regulated clouds. The Risk Reduction Agent, also in preview, coordinates remediation across Microsoft, AWS and Google Cloud and can recommend vulnerability shielding: targeted, temporary protection when patches cannot be deployed immediately.

Microsoft is extending identity and data controls to agents in customer environments. Discovery in Agent 365 and Entra assigns identities to discovered agents for identity governance. Intune policies can contain supported local agents on devices, while Defender, Purview Data Loss Prevention and Conditional Access govern runtime access and sensitive data. Discovery includes third-party cloud agents on AWS and Google Cloud, as well as 60 local agent types, including Claude Code, OpenAI Codex and GitHub Copilot.

Share:

Comments (0)

Leave a comment

No comments yet. Be the first!