Downgrading Microsoft Information Protection (MIP) labels

This section cannot be readily mitigated at a system level with preventive controls since it is based on the abuse of fundamental features of the system.

An agent capable of Endpoint Detection and Response (EDR) is a software agent installed on organization endpoints (such as laptops and servers) that (at a minimum) records the Operating System, application, and network activity on an endpoint.

Typically EDR operates in an agent/server model, where agents automatically send logs to a server, where the server correlates those logs based on a rule set. This rule set is then used to surface potential security-related events, that can then be analyzed.

An EDR agent typically also has some form of remote shell capability, where a user of the EDR platform can gain a remote shell session on a target endpoint, for incident response purposes. An EDR agent will typically have the ability to remotely isolate an endpoint, where all network activity is blocked on the target endpoint (other than the network activity required for the EDR platform to operate).

An agent capable of User Activity Monitoring (UAM) is a software agent installed on organization endpoints (such as laptops); typically, User Activity Monitoring agents are only deployed on endpoints where a human user Is expected to conduct the activity.

The User Activity Monitoring agent will typically record Operating System, application, and network activity occurring on an endpoint, with a focus on activity that is or can be conducted by a human user. The purpose of this monitoring is to identify undesirable and/or malicious activity being conducted by a human user (in this context, an Insider Threat).

Typical User Activity Monitoring platforms operate in an agent/server model where activity logs are sent to a server for automatic correlation against a rule set. This rule set is used to surface activity that may represent Insider Threat related activity such as capturing screenshots, copying data, compressing files or installing risky software.

Other platforms providing related functionality are frequently referred to as User Behaviour Analytics (UBA) platforms.

An agent capable of User Behaviour Analytics (UBA) is a software agent installed on organizational endpoints (such as laptops). Typically, User Activity Monitoring agents are only deployed on endpoints where a human user is expected to conduct the activity.

The User Behaviour Analytics agent will typically record Operating System, application, and network activity occurring on an endpoint, focusing on activity that is or can be conducted by a human user. Typically, User Behaviour Analytics platforms operate in an agent/server model where activity logs are sent to a server for automatic analysis. In the case of User Behaviour Analytics, this analysis will typically be conducted against a baseline that has previously been established.

A User Behaviour Analytic platform will typically conduct a period of ‘baselining’ when the platform is first installed. This baselining period establishes the normal behavior parameters for an organization’s users, which are used to train a Machine Learning (ML) model. This ML model can then be later used to automatically identify activity that is predicted to be an anomaly, which is hoped to surface user behavior that is undesirable, risky, or malicious.

Other platforms providing related functionality are frequently referred to as User Activity Monitoring (UAM) platforms.

Microsoft Information Protection (MIP) sensitivity labels are metadata-based security attributes applied to files, emails, and other content within Microsoft 365 environments. MIP sensitivity labels act as a form of document-centric access control, embedding security policies directly into files and emails. By tagging content with persistent metadata that enforces encryption, access restrictions, and visual markings, MIP labels ensure that data protection travels with the document—regardless of where it's stored or shared—providing consistent security across organizational and cloud boundaries. 

MIP labels are centrally defined through the Microsoft Purview compliance portal and persist within the content itself—stored in metadata streams such as Office document custom properties or XML parts. Labels can be applied manually by users or automatically via content inspection rules, data classification policies, or machine learning models. Once applied, labels can enforce a range of protections, including Azure Information Protection (AIP)-based encryption, visual markings (e.g., headers, footers, watermarks), and access restrictions.

Because MIP labels are integrated with Microsoft 365 applications and services, they serve as a powerful mechanism for monitoring and auditing sensitive data handling. Labeling events generate detailed telemetry that can help identify suspicious or non-compliant user behavior, such as:

• Downgrading a file from a more restrictive label (e.g., "Highly Confidential") to a less restrictive one (e.g., "Public") before exfiltration.
• Applying inconsistent labels to similar types of content.
• Bypassing automatic labeling recommendations or ignoring mandatory labeling prompts.
• Accessing or modifying labeled content outside normal working hours or from anomalous locations.

Detection can be implemented across various Microsoft platforms:

• Microsoft Purview (formerly Microsoft 365 Compliance Center) provides audit logs and activity explorer views for label application, modification, and removal.
• Microsoft Defender for Cloud Apps (MCAS) enables near real-time monitoring of MIP label usage across Microsoft 365 and integrated third-party services.
• Microsoft Sentinel can ingest logs from Microsoft Purview, Azure AD, and Microsoft Defender to correlate labeling activity with other insider threat signals.
• Microsoft Defender for Endpoint monitors endpoint behavior, which can be used to identify lateral movement, data access anomalies, or unauthorized label downgrades.
•  

Detection rules can be enriched with user and entity behavior analytics (UEBA), data loss prevention (DLP) events, and identity-based risk signals (e.g., unusual sign-ins or privilege escalations) to increase fidelity and reduce false positives.