Microsoft Purview – Data Security Posture Management (DSPM) for AI
Introduction to DSPM for AI
In an age where Artificial Intelligence (AI) is rapidly transforming industries, ensuring the security and compliance of AI integrations is paramount. Microsoft Purview Data Security Posture Management (DSPM) for AI helps organizations monitor AI activity, enforce security policies, and prevent unauthorised data exposure.
Microsoft Purview Data Security Posture Management (DSPM) for AI addresses three primary areas: Recommendations, Reports, and Data Assessments. DSPM for AI assists in identifying vulnerabilities associated with unprotected data and enables prompt action to enhance data security posture and mitigate risks effectively.
Getting Started with DSPM for AI
To manage and mitigate AI-related risks, Microsoft Purview provides easy-to-use graphical tools and comprehensive reports. These features allow you to quickly gain insights into AI use within your organization. The one-click policies offered by Microsoft Purview simplify the process of protecting your data and ensuring compliance with regulatory requirements.
Prerequisites for Data Security Posture Management for AI
To use DSPM for AI from the Microsoft Purview portal or the Microsoft Purview compliance portal, you must have the following prerequisites:
You have the right permissions.
Monitoring Copilot interactions requires:
Users are assigned a license for Microsoft 365 Copilot.
o Microsoft Purview auditing enabled. Check instructions for Turn auditing on or off.
Required for monitoring interactions with third-party generative AI sites:
Devices are onboarded to Microsoft Purview, required for:
Gaining visibility into sensitive information that's shared with third-party generative AI sites. (e.g., credit card numbers pasted into ChatGPT).
Applying endpoint DLP policies to warn or block users from sharing sensitive information with third-party generative AI sites. (e.g. a user identified as elevated risk in Adaptive Protection is blocked with the option to override when they paste credit card numbers into ChatGPT)
The Microsoft Purview browser extension is deployed to users and required to discover site visits to third-party generative AI sites.
Things to consider
Recommendations may differ based on M365 licenses and features.
Not all recommendations are relevant for every tenant and can be dismissed.
Any default policies created while Data Security Posture Management for AI was in preview and named Microsoft Purview AI Hub won't be changed. For example, policy names will retain their Microsoft AI Hub -prefix.
In this blog post we are going to focus on Recommendations.
Recommendations
Let's explore each of the recommendations in detail, which will encompass one-click policy creation, data assessments, step-by-step guidance, and regulations. The data in the reports section will be contingent upon the completion of each recommendation.
Figure 1: Recommendations – DSPM for AI
Control unethical behaviour in AI
Type: One-click policy
Solution: Communication Compliance
Description: This policy identifies sensitive information within prompts and response activities in Microsoft 365 Copilot.
Action:
Create policy to setup a one-click policy.
Conditions: Content matches any of these trainable classifiers: Regulatory Collusion, Stock manipulation, Unauthorized disclosure, Money laundering, Corporate Sabotage, Sexual, Violence, Hate, Self-harm
By default, all users and groups are added.
The customisation of the policy is also available during the one-click policy creation process.
Figure 2: Recommendations – One-click policy
Guided assistance to AI regulations
Type: New AI regulations
Solution: Compliance manager
Description: This recommendation is based on the NIST AI RMF regulations, suggesting actions to help users protect data during interactions with AI systems.
Action:
Monitor AI interaction logs: Go to Audit logs, configure search with workload filter, select copilot and sensitive information type and review search results.
Monitor AI interactions in other AI apps: Navigate to DSPM for AI and review interactions in other AI apps for sensitive content and turn on policies to discover data across AI interactions and other AI apps.
Flag risky communication and content in AI interactions: Create Communication compliance policy to define the necessary conditions and fields and select Microsoft Copilot as location.
Prevent sensitive data from being shared in AI apps: Create Data loss prevention (DLP) policy with sensitive information type as conditions for Teams and Channel messages location.
Manage retention and deletion policies for AI interactions: Create a retention policy for Teams chat and Microsoft 365 Copilot interactions to preserve relevant AI activities for a longer duration while promptly deleting non-relevant user actions.
Protect sensitive data referenced in Copilot responses
Type: Assessment
Solution: Data assessments
Description: Use data assessments to identify potential oversharing risks, including unlabelled files.
Action:
Create Data Assessments, Navigate to DSPM for AI - Data Assessments and Create Assessments.
Enter assessment name and description
Select users and data sources to assets for oversharing data
Conduct the assessment scan and review the results to gain insights into oversharing risks and recommended solutions to restrict access to sensitive data.
Implement the necessary fixes to protect your data.
Discover and govern interactions with ChatGPT Enterprise AI (preview)
Type: ChatGPT Enterprise AI (Data discovery)
Solution: Microsoft Purview Data Map
Description: Register ChatGPT Enterprise workspace to discover and govern interactions with ChatGPT Enterprise AI.
Action:
If you’re organisation is using ChatGPT Enterprise, then enable the Connector
In Microsoft Azure, use Key Vault to manage credentials for third-party connectors: Use Key Vault to create and manage the secret for the ChatGPT Enterprise AI Connector.
In Microsoft Purview, configure the new connector using Data Map: How to manage data sources in the Microsoft Purview Data Map
Create and start a new scan: Create a new scan, select credential, review, and run the scan.
Protect sensitive data referenced in Microsoft 365 Copilot (preview)
Type: Data Security
Solution: Data loss prevention
Description: Content with sensitivity labels will be restricted from Copilot interactions with a data loss prevention policy.
Action:
Create a custom DLP policy and select Microsoft 365 Copilot as the data source.
Create a custom rule
o Condition: content contains sensitivity labels.
o Action: Prevent Copilot from processing content.
Figure 3: Custom DLP policy condition and action
Fortify your data security
Type: Data security
Solution: Data loss prevention
Description: Data security risks can range from accidental oversharing of information outside of the organization to data theft with malicious intent. These policies will protect against the data security risks with AI apps.
Action:
A one-click policy is available to create a data loss prevention (DLP) policy for endpoints (devices), aimed at blocking the transmission of sensitive information to AI sites.
It utilises Adaptive Protection to give a warn-with-override alert to users with elevated risk levels who attempt to paste or upload sensitive information to other AI assistants in browsers such as Edge, Chrome, and Firefox. This policy covers all users and groups in your org in test mode.
Figure 4: Block with override for elevated risk users
Information Protection Policy for Sensitivity Labels
Type: Data security
Solution: Sensitivity Labels
Description: This policy will set up default sensitivity labels to preserve document access rights and protect Microsoft 365 Copilot output.
Action:
Create policies will navigate to Information protection portal to set up sensitivity labels and publishing policy.
Protect your data from potential oversharing risks
Type: Data Security
Solution: Data Assessment
Description: Data assessments provide insights on potential oversharing risks within your organisation for SharePoint Online and OneDrive for Business (roadmap) along with fixes to limit access to sensitive data. This report will include sharing links.
Action:
This is a default oversharing assessment policy.
To see the latest oversharing scan results:
Select View latest results and choose a data source.
Complete fixes to secure your data.
Figure 5: Data assessments – Oversharing assessment data with sharing links report
Use Copilot to improve your data security posture (preview)
Type: Data security posture management
Solution: Data security posture management (DSPM)
Description: Data Security Posture Management (preview) combines deep insights with Security Copilot capabilities to help you identify and address security risks in your org.
Benefits:
Data security recommendations
Gain insights into your data security posture and get recommendations protecting sensitive data and closing security gaps.
Data security trends
Track your org's data security posture over time with reports summarizing sensitive label usage, DLP policy coverage, changes in risky user behaviour, and more.
Security Copilot
Security Copilot helps you investigate alerts, identify risk patterns, and pinpoint the top data security risks in your org.
