[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fUdn4hPI9704-s_6bTTV6en3iGAWLmdEz4XoGIZ50nO4":3},{"lesson":4},{"id":5,"slug":6,"article_id":7,"title":8,"body":9,"prevention":10,"framework_refs":11,"framework_notes":25,"status":39,"created_at":40,"published_at":41,"article":42,"tags":46,"podcasts":65,"controls":66},"17bb7200-d654-4c42-acff-bed73f8ba809","microsoft-bolsters-ai-agent-security-with-dlp-auto-labeling-and-copilot-controls","ec9670aa-10fc-4ff2-a22f-5d7ec96af171","Microsoft Bolsters AI Agent Security with DLP, Auto-Labeling, and Copilot Controls","As AI agents become deeply embedded in enterprise workflows across devices and cloud platforms, organizations face new risks around sensitive data being inadvertently exposed to or processed by AI tools. Microsoft's September 2026 updates highlight that without proper data loss prevention and access controls, confidential information can leak through AI interactions in ways traditional security controls were never designed to catch. The expansion of auto-labeling and eDiscovery for Copilot-generated content underscores that AI-produced data carries the same compliance obligations as human-created content. Failing to govern AI agent behavior and data flows leaves organizations exposed to regulatory penalties and reputational harm.","**Immediate actions:**\n- Enable Microsoft Purview DLP policies specifically targeting sensitive data categories transmitted to or processed by AI tools and Copilot applications.\n- Apply auto-labeling rules to AI-generated content to ensure it inherits the same classification and protection as equivalent human-created data.\n- Audit current Microsoft Entra Global Secure Access configurations to ensure AI agent traffic is subject to the same access and inspection policies as standard user traffic.\n\n**Long-term improvements:**\n- Build a comprehensive AI asset inventory that tracks all deployed AI agents, their data access scopes, and associated cloud and device integrations.\n- Establish governance policies defining which data classifications may be shared with AI tools, and enforce these boundaries through technical controls rather than user awareness alone.\n- Integrate AI-related eDiscovery and archiving workflows into existing records management and compliance programs to avoid blind spots in legal hold processes.\n\n**Detection measures:**\n- Configure Security Copilot investigation summaries and alerts to surface anomalous AI data-access patterns for rapid triage by the security operations team.\n- Implement continuous monitoring of DLP policy match logs for AI-directed data flows and schedule quarterly reviews to tune policies as AI usage evolves.",[12,13,14,15,16,17,18,19,20,21,22,23,24],"NIST AI RMF – Govern 1.1, Map 1.5","NIST SP 800-53 AC-4 (Information Flow Enforcement)","NIST SP 800-53 SC-28 (Protection of Information at Rest)","NIST SP 800-53 SI-12 (Information Management and Retention)","CIS Control 3 – Data Protection","CIS Control 6 – Access Control Management","CIS Control 8 – Audit Log Management","GDPR Article 5(1)(f) – Integrity and Confidentiality","GDPR Article 25 – Data Protection by Design and by Default","GDPR Article 30 – Records of Processing Activities","ISO\u002FIEC 27001:2022 – A.5.12 Classification of Information","ISO\u002FIEC 27001:2022 – A.5.33 Protection of Records","ITIL 4 – Service Configuration Management Practice",{"CIS Control 3 – Data Protection":26,"NIST AI RMF – Govern 1.1, Map 1.5":27,"CIS Control 8 – Audit Log Management":28,"CIS Control 6 – Access Control Management":29,"NIST SP 800-53 AC-4 (Information Flow Enforcement)":30,"ISO\u002FIEC 27001:2022 – A.5.33 Protection of Records":31,"GDPR Article 30 – Records of Processing Activities":32,"ITIL 4 – Service Configuration Management Practice":33,"GDPR Article 5(1)(f) – Integrity and Confidentiality":34,"NIST SP 800-53 SC-28 (Protection of Information at Rest)":35,"ISO\u002FIEC 27001:2022 – A.5.12 Classification of Information":36,"NIST SP 800-53 SI-12 (Information Management and Retention)":37,"GDPR Article 25 – Data Protection by Design and by Default":38},"Data protection controls such as labelling and DLP limit sensitive data exposure through AI tools.","AI governance and context mapping help organisations decide what data AI agents may access and how risks are managed.","Logging AI interactions supports investigation and eDiscovery when sensitive data is exposed.","Access control ensures AI agents only surface data the user is allowed to see.","Information flow enforcement, such as DLP for AI interactions, prevents confidential data leaking into prompts or outputs.","Protecting records, including AI-generated content, ensures they are retained and discoverable.","Records of processing should include AI tools that handle personal data.","Configuration management tracks AI tool settings and policies so controls stay consistent.","Integrity and confidentiality require protecting personal data processed by AI tools against leakage.","Protecting stored data, including AI-generated content, ensures sensitive information remains encrypted and access-controlled.","Classifying information, including via auto-labelling, lets DLP and AI controls treat it correctly.","Managing retention of Copilot-generated content ensures it follows the same rules as other records.","Building DLP and labelling into AI deployments by default is data protection by design.","published","2026-09-24T20:22:23.386761+00:00","2026-09-24T20:22:23.286+00:00",{"id":7,"url":43,"slug":44,"title":45},"https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fsecurity\u002Fblog\u002F2026\u002F09\u002F24\u002Fwhats-new-in-microsoft-security-september-2026\u002F","what-s-new-in-microsoft-security-september-2026-f9fa63","​​​​​​​​What’s new in Microsoft Security: September 2026​​",[47,53,59],{"id":48,"name":49,"slug":50,"description":51,"color":52},"1ec88fde-2d0f-4ed8-932a-33f5ccc0fdc7","Access Control","access-control","Excessive privileges, missing MFA, weak auth","#f97316",{"id":54,"name":55,"slug":56,"description":57,"color":58},"859cf0ad-a7e9-42bb-a75d-bac6511fa5d5","Configuration Management","configuration-management","Misconfigs, default credentials, exposed services","#eab308",{"id":60,"name":61,"slug":62,"description":63,"color":64},"c8b843a5-d5a7-41d1-8d3b-cabded09d2ef","Data Protection","data-protection","Unencrypted data, missing DLP, poor classification","#3b82f6",[],[67,71,75],{"slug":68,"name":69,"problem":70},"zero-trust-access","Zero trust network access (ZTNA)","Remote access no longer means exposing VPN appliances to the internet.",{"slug":72,"name":73,"problem":74},"data-loss-prevention","Data loss prevention (DLP)","Sensitive data cannot quietly leave by email, upload, USB stick or AI tool.",{"slug":76,"name":77,"problem":78},"ai-security","AI application security","Prompt injection, data leaks through AI tools and unapproved AI use get caught."]