Secure AI use with DSPM AI in Microsoft Purview DSPM: setup tasks that give DSPM signals from Copilot, enterprise AI apps and other AI apps, Apps and agents with policy coverage per app, app details with risky AI usage, AI observability for agents, the objective Prevent exfiltration to unsanctioned AI apps and its remediation plan with a DLP policy, an Insider Risk Management policy and a Recall rule, the difference between browser and network signals, and the AI Usage and Risk report. W DATA SECURITY POSTURE MANAGEMENT Secure AI use with DSPM Ada Admin Eli Employee AI IN DSPM Three kinds of AI, three signals COPILOT AND AGENTS Microsoft 365 Copilot,Copilot Studio and agents. SignalAudit and collectionpolicies. ENTERPRISE AI APPS AI apps your organizationhas connected, such asChatGPT Enterprise. SignalCollection policies andpay-as-you-go billing. OTHER AI APPS AI sites people open in abrowser, like ChatGPT,Gemini or Claude. SignalBrowser extension ononboarded devices, or network. No signal, no insight. Setup decides what DSPM can see. BROWSER OR NETWORK Two ways to see other AI apps BROWSER The Purview browser extension ononboarded devices sees what peoplepaste into AI sites. ✓ In place here NETWORK A SASE or SSE provider connected toPurview sees AI traffic from apps,not only from the browser. Not set up A quiet objective can mean no signal. Check the setup tasks before you trust a green status. Microsoft Purview Setup tasks These setup tasks provide visibility to sensitive data, including risks and security controls to mitigate those risks. Not Started1 Dismissed0 Completed9 ↻ Refresh10 items Setup taskTypeDurationCompleted byCompleted on Apps and agents (preview) Understand the depth and breadth of Microsoft Purview protection for AI applications and agents in the last 30 days. AppsAgents ↻ Refresh⤓ Export24 items AI appProtection statusUser trendData protectionData complianceAgents ◆ Microsoft Copilot StudioMicrosoft Copilot Studio ✓ Monitored Data security and compliance policies monitoring this AI app Data protection policies0 Data compliance policies0 Alerts for risky interactions in Copilot experiences and agents Generates alerts for risky interactions and updates the user risk level. Users with risky AI usageLast 30 days Received sensitive responses30 Entered risky prompts23 User risk and activity User risk distribution 2 No risk AI observability Get a centralized view of agent activity across your organization. Cloud agents (587)Local agents (0)AI apps (3) Key metrics 587 cloud agents AvailableBlockedDeleted 58403 ↑ 0.85% No insights yetData is currently unavailable 0 risky interactions OversharingExfiltrationUnethical 000 NamePlatformStatusRisk levelRisk types Data security objectives ! PData protection Prevent exfiltration to unsanctioned AI apps 5 sensitive information were shared in unsanctioned AI apps in the last 29 days, an increase of 400%. This could indicate emerging data exfiltration patterns. 5 sensitive informationexfiltrated to AI apps ↑ +400% in last 29 days 1Users exfiltrating 5 ↗ +400%Unique interactions 1Unsanctioned AI apps Take action to reduce risk Taking action could reduceexfiltrated sensitiveinformation from 5 to 0,about a 100% decrease. View remediation planCopilot prompts View objective → AI Usage & Risk Copilot experiences & agents Enterprise AI apps Other AI apps Activity Remediation plan: Prevent exfiltration to unsanctioned AI apps Increased risks detected in the last 30 days Sensitive data is being shared with unsanctioned AI applications. Estimated impact on risk pattern 18.79.40 Today Sep 20Sep 27Oct 4Oct 11Oct 18 Data exfiltration trends Prevent exfiltration to AI apps with DLP Policy nameDSPM - Prevent exfiltration to unsanctioned AI apps Data to detect (DLP rules)Content containing SITs found in exfiltration risk patterns Apply policy to these data sourcesBrowser, Network Policy modeSimulation mode Policy created inData Loss Prevention Next stepsTo begin collecting network traffic in Microsoft Purview, you mustfirst configure one or more Secure Access Service Edge (SASE)or Security Service Edge (SSE) integrations. Clear all Apply Cancel Good to know 1/7 DATA SECURITY POSTURE MANAGEMENT Secure AI use with DSPM DSPM shows how people use Copilot, agents and other AI apps. We check what DSPM can see, what is in use, and turn an AI objective into a DLP policy. informasjonsbeskyttelse.no