Exfiltration via Generative AI Platform

A Data Loss Prevention (DLP) solution refers to policies, technologies, and controls that prevent the accidental and/or deliberate loss, misuse, or theft of data by members of an organization. Typically, DLP technology would take the form of a software agent installed on organization endpoints (such as laptops and servers).

Typical DLP technology will alert on the potential loss of data, or activity which might indicate the potential for data loss. A DLP technology may also provide automated responses to prevent data loss on a device.

Domain Name System (DNS) filtering allows the blocking of domain resolution for specific domains or automatically categorized classes of domains (depending on the functionality of the software or appliance being used). DNS filtering prevents users from accessing blocked domains, regardless of the IP address the domains resolve to.

Examples of automatically categorized classes of domains are ‘gambling’ or ‘social networking’ domains. Automatic categorizations of domains are typically conducted by the software or appliance being used, whereas specific domains can be blocked manually. Most DNS filtering software or appliances will provide the ability to use Regular Expressions (RegEx) to (for example) also filter all subdomains on a specified domain.

DNS filtering can be applied on an individual host, such as with the hosts file, or for multiple hosts via a DNS server or firewall.

An Acceptable Use Policy (AUP) is a set of rules outlining acceptable and unacceptable uses of an organization's computer systems and network resources. It acts as a deterrent to prevent employees from conducting illegitimate activities by clearly defining expectations, reinforcing legal and ethical standards, establishing accountability, specifying consequences for violations, and promoting education and awareness about security risks.

An enterprise-managed browser is a web browser controlled by an organization to enforce security policies, manage employee access, and ensure compliance. It allows IT administrators to monitor and restrict browsing activities, apply security updates, and integrate with other enterprise tools for a secure browsing environment.

Microsoft Information Protection (MIP) sensitivity labels are powerful tools for preventing unauthorized access, data leakage, data loss and other types of insider events through classification and protection of sensitive content. When applied to documents, emails, and other content, MIP labels embed metadata that enforces encryption, access control policies, and usage restrictions — all of which persist even if the content is shared or moved outside the organization’s environment. This proactive protection mechanism helps to ensure that data loss, misuse, or regulatory breaches are minimized, regardless of where or how the data is accessed.

Persistent Protection through Azure Rights Management (Azure RMS)
One of the key features of MIP labels is their ability to enforce encryption and access control via Azure Rights Management (Azure RMS). When a document or email is assigned a sensitivity label such as Highly Confidential, it triggers policies that encrypt the file, limiting who can open it and what actions can be performed. For example, a Highly Confidential document might be encrypted so that only authorized users in specific security groups can access it. Additionally, these policies may prevent recipients from forwarding, printing, copying, or even accessing the document offline, ensuring that sensitive data cannot be shared beyond authorized channels.

Automatic and Recommended Labeling
MIP labels also support automatic and recommended labeling. Labels can be automatically applied based on content that is identified as sensitive (such as credit card numbers, Social Security numbers, or intellectual property). This reduces reliance on users to manually select the correct label, ensuring that content is always classified according to its sensitivity level. For example, a file containing financial data or personally identifiable information (PII) may automatically receive a Confidential label, which immediately triggers encryption and access controls. By applying labels automatically, organizations can minimize the risk of human error in classifying sensitive content and ensure that protective measures are consistently applied.

Enforcing Access Governance and User Restrictions
MIP labels are directly integrated with Azure Active Directory (Azure AD) and Microsoft 365 security groups, allowing organizations to enforce access governance. Each label can define the users or groups who are permitted to access certain types of content. For example, a document labeled Confidential may be restricted to a specific department or team, preventing unauthorized users from viewing or editing it. Access to content labeled with higher sensitivity, such as Highly Confidential, can be further restricted to executives or security professionals, ensuring that only authorized individuals can access critical business data. These policies persist even when the content is shared outside the organization or accessed on non-corporate devices.

Blocking Unauthorized Sharing and Transfers
Through integration with Microsoft Defender for Office 365 and Data Loss Prevention (DLP) policies, MIP labels can enforce automatic blocking of unauthorized sharing or transfer of sensitive content. For example, when a document is labeled as Internal Use Only, any attempt to share it externally via email, cloud storage, or external USB devices can be blocked automatically by DLP policies. Labels can also be configured to restrict sharing links to specific people or groups and can enforce expiration on shared links. This ensures that sensitive data remains within the organization and cannot be accessed by unauthorized individuals or systems.

Policy Enforcement in Microsoft Teams and SharePoint
MIP labels are integrated across key collaboration tools like Microsoft Teams and SharePoint, providing seamless protection in the cloud. In these environments, sensitivity labels govern sharing permissions, access rights, and file handling. For instance, if a file is labeled as Confidential, it might be restricted from being shared externally via Teams or SharePoint. These platforms can also prevent file download and sharing for users in unmanaged or non-compliant environments, ensuring that sensitive data cannot be accessed outside the organization's controlled infrastructure. MIP labels also enable policies that enforce restrictions on guest access, preventing external parties from viewing or editing sensitive content unless explicitly permitted.

Blocking Label Downgrades and Enforcing Label Change Justification
To prevent unauthorized downgrading of content labels, MIP provides mechanisms to block label downgrades without proper justification. For example, a user may not be allowed to change a document’s label from Confidential to Public without providing an explicit justification. Such actions are logged and may trigger alerts for review by security teams. This ensures that users cannot bypass sensitive information protection policies by reclassifying content to a lower sensitivity level. Moreover, any label changes are auditable, helping organizations track and monitor potential attempts to circumvent security protocols.

Preventing Exfiltration in Cloud and Endpoint Contexts
MIP labels integrate with Microsoft Defender for Endpoint and Defender for Cloud Apps, providing protection against exfiltration of sensitive data through cloud and endpoint channels. By applying labels to sensitive documents, organizations can enforce controls that restrict their movement across corporate boundaries. For example, when a file labeled Confidential is accessed from an unmanaged device or through a risky application, it may be blocked from being downloaded or printed, preventing potential exfiltration. Additionally, organizations can configure conditional access policies to prevent data access based on the device’s compliance or security status, ensuring that sensitive information is protected even when users access it from external sources.

Training should equip employees to recognize manipulation tactics, such as social engineering and extortion, that are used to coerce actions and behaviors harmful to the individual and/or the organization. The training should also encourage and guide participants on how to safely report any instances of coercion.

A web proxy can allow for specific web resources to be blocked, preventing clients from successfully connecting to them.

Provide a process for all staff members to report concerning and/or suspicious behaviour to the organization's security team for review. An internal whistleblowing process should take into consideration the privacy of the reporter and the subject(s) of the report, with specific regard to safeguarding against reprisals against reporters.

Establish and maintain formal, well-communicated pathways for personnel to request resources, report deficiencies, or propose operational improvements. By providing structured mechanisms to meet legitimate needs, organizations reduce the likelihood that subjects will bypass policy controls through opportunistic or unauthorized actions.

Implementation Approaches

• Create clear, accessible request processes for technology needs, system enhancements, and operational support requirements.
• Ensure personnel understand how to escalate unmet needs when standard processes are insufficient, including rapid escalation pathways for operational environments.
• Maintain service-level agreements (SLAs) or expected response times to requests, ensuring perceived barriers or delays do not incentivize unofficial action.
• Integrate feedback mechanisms that allow users to suggest improvements or report resource shortfalls anonymously or through designated representatives.
• Publicize successful examples where formal channels resulted in legitimate needs being met, reinforcing the effectiveness and trustworthiness of the system.

Operational Principles

• Responsiveness: Requests must be acknowledged and processed promptly to prevent frustration and informal workarounds.
• Transparency: Personnel should be informed about request status and outcomes to maintain trust in the process.
• Accountability: Ownership for handling requests must be clearly assigned to responsible teams or individuals.
• Cultural Integration: Leaders and supervisors should reinforce the use of formal channels and discourage unsanctioned self-remediation efforts.

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.

A Data Loss Prevention (DLP) solution refers to policies, technologies, and controls that prevent the accidental and/or deliberate loss, misuse, or theft of data by members of an organization. Typically, DLP technology would take the form of a software agent installed on organization endpoints (such as laptops and servers).

Typical DLP technology will alert on the potential loss of data, or activity which might indicate the potential for data loss. A DLP technology may also provide automated responses to prevent data loss on a device.

Monitor outbound DNS traffic for unusual or suspicious queries that may indicate DNS tunneling. DNS monitoring entails observing and analyzing Domain Name System (DNS) queries and responses to identify abnormal or malicious activities. This can be achieved using various security platforms and network appliances, including Network Intrusion Detection Systems (NIDS), specialized DNS services, and Security Information and Event Management (SIEM) systems that process DNS logs.

Commercial security software may have the ability to generate alerts when suspected tampering is detected, such as interacting with the process in memory, or attempting to access files related to its operation.

Depending on the solution used, web proxies can provide a wealth of information about web-based activity. This can include the IP address of the system making the web request, the URL requested, the response code, and timestamps.

An organization must perform SSL/TLS interception to receive the most complete information about these connections.