Work IQ: A2A and MCP long running tasks

🚨 The Signal: New capabilities enable long-running automated tasks for Copilot and other services. This introduces new vectors for persistent malicious activity and data exfiltration, requiring enhanced monitoring and governance.

The Impact

Automated processes and Copilot agents are affected, increasing risk of persistent threats and data exfiltration.

• Security Teams: New persistent task vectors increase monitoring complexity.
• Copilot Admins: Need to review and secure long-running agent tasks.
• Data Owners: Increased risk of data exfiltration via automated processes.
• Compliance Teams: New audit trails for long-running tasks require review.

The Action

1. Review Microsoft Purview audit logs for long-running A2A/MCP tasks.
2. Implement Conditional Access policies to restrict A2A/MCP task initiation.
3. Configure Microsoft Defender for Cloud Apps policies to monitor A2A/MCP task activity.
4. Establish data loss prevention (DLP) policies for data accessed by A2A/MCP tasks.
5. Regularly review Copilot agent permissions and task definitions.

Domain: Agentic-AI · Impact: high · Workload: Other