Issue with the Canadian Drivers License SIT
Did any face an issue with the Canadian Driver's License SIT in the DLP policy? We see a lot of false positives especially around BC province number, NB, PrinceEdward Island and Saskatchewan. These provinces has just digits which can flag any kind of digits. Even if we use some custom RegEx and reduced keyword list, it still flags a lot of false positives. We see this as more and more customers are not happy with it. Has anyone found a breakthrough or best solution for deploying the Canadian Driver's License DLP?Search and Purge workflow in the new modern eDiscovery experience
With the retirement of Content Search (Classic) and eDiscovery Standard (Classic) in May, and alongside the future retirement of eDiscovery Premium (Classic) in August, organizations may be wondering how this will impact their existing search and purge workflow. The good news is that it will not impact your organizations ability to search for and purge email, Teams and M365 Copilot messages; however there are some additional points to be careful about when working with purge with cmdlet and Graph alongside of the modern eDiscovery experience. We have made some recent updates to our documentation regarding this topic to reflect the changes in the new modern eDiscovery experience. These can be found below and you should ensure that you read them in full as they are packed with important information on the process. Find and delete email messages in eDiscovery | Microsoft Learn Find and delete Microsoft Teams chat messages in eDiscovery | Microsoft Learn Search for and delete Copilot data in eDiscovery | Microsoft Learn The intention of this first blog post in the series is to cover the high-level points including some best practices when it comes to running search and purge operations using Microsoft Purview eDiscovery. Please stay tuned for further blog posts intended to provide more detailed step-by-step of the following search and purge scenarios: Search and Purge email and Teams messages using Microsoft Graph eDiscovery APIs Search and Purge email messages using the Security and Compliance PowerShell cmdlets I will update this blog post with the subsequent links to the follow-on posts in this series. So let’s start by looking at the two methods available to issue a purge command with Microsoft Purview eDiscovery, they are the Microsoft Graph eDiscovery APIs or the Security and Compliance PowerShell cmdlets. What licenses you have dictates which options are available to you and what type of items you can be purge from Microsoft 365 workloads. For E3/G3 customers and cases which have the premium features disabled You can only use the PowerShell cmdlets to issue the purge command You should only purge email items from mailboxes and not Teams messages You are limited to deleting 10 items per location with a purge command For E5/G5 customers and cases which have the premium features enabled You can only use the Graph API to issue the purge command You can purge email items and Teams messages You can delete up to 100 items per location with a purge command To undertake a search and then purge you must have the correct permissions assigned to your account. There are two key Purview Roles that you must be assigned, they are: Compliance Search: This role lets users run the Content Search tool in the Microsoft Purview portal to search mailboxes and public folders, SharePoint Online sites, OneDrive for Business sites, Skype for Business conversations, Microsoft 365 groups, and Microsoft Teams, and Viva Engage groups. This role allows a user to get an estimate of the search results and create export reports, but other roles are needed to initiate content search actions such as previewing, exporting, or deleting search results. Search and Purge: This role lets users perform bulk removal of data matching the criteria of a search. To learn more about permissions in eDiscovery, along with the different eDiscovery Purview Roles, please refer to the following Microsoft Learn article: Assign permissions in eDiscovery | Microsoft Learn By default, eDiscovery Manager and eDiscovery Administrators have the “Compliance Search” role assigned. For search and purge, only the Organization Management Purview Role group has the role assigned by default. However, this is a highly privileged Purview Role group and customers should considering using a custom role group to assign the Search and Purge Purview role to authorised administrators. Details on how to create a custom role group in Purview can be found in the following article. Permissions in the Microsoft Purview portal | Microsoft Learn It is also important to consider the impact of any retention policies or legal holds will have when attempting to purge email items from a mailbox where you want to hard delete the items and remove it completely from the mailbox. When a retention policy or legal hold is applied to a mailbox, email items that are hard deleted via the purge process are moved and retained in the Recoverable Items folder of the mailbox. There purged items will be retained until such time as all holds are lifted and until the retention period defined in the retention policy has expired. It is important to note that items retained in the Recoverable Items folder are not visible to users but are returned in eDiscovery searches. For some search and purge use cases this is not a concern; if the primary goal is to remove the item from the user’s view then additional steps are required. However if the goal is to completely remove the email item from the mailbox in Exchange Online so it doesn't appear in the user’s view and is not returned by future eDiscovery searches then additional steps are required. They are: Disable client access to the mailbox Modify retention settings on the mailbox Disable the Exchange Online Managed Folder Assistant for the mailbox Remove all legal holds and retention policies from the mailbox Perform the search and purge operation Revert the mailbox to its previous state These steps should be carefully followed as any mistake could result in additional data that is being retained being permanently deleted from the service. The full detailed steps can be found in the following article. Delete items in the Recoverable Items folder mailboxes on hold in eDiscovery | Microsoft Learn Now for some best practice when running search and purge operations: Where possible target the specific locations containing the items you wish to purge and avoid tenant wide searches where possible If a tenant wide search is used to initially locate the items, once the locations containing the items are known modify the search to target the specific locations and rerun the steps Always validate the item report against the statistics prior to issuing the purge command to ensure you are only purging items you intend to remove If the item counts do not align then do not proceed with the purge command Ensure admins undertaking search and purge operations are appropriately trained and equipped with up-to-date guidance/process on how to safely execute the purge process The search conditions Identifier, Sensitivity Label and Sensitive Information Type do not support purge operations and if used can cause un-intended results Organizations with E5/G5 licenses should also take this opportunity to review if other Microsoft Purview and Defender offerings can help them achieve the same outcomes. When considering the right approach/tool to meet your desired outcomes you should become familiar with the following additional options for removing email items: Priority Clean-up (link): Use the Priority cleanup feature under Data Lifecycle Management in Microsoft Purview when you need to expedite the permanent deletion of sensitive content from Exchange mailboxes, overriding any existing retention settings or eDiscovery holds. This process might be implemented for security or privacy in response to an incident, or for compliance with regulatory requirements. Threat Explorer (link): Threat Explorer in Microsoft Defender for Office 365 is a powerful tool that enables security teams to investigate and remediate malicious emails in near real-time. It allows users to search for and filter email messages based on various criteria - such as sender, recipient, subject, or threat type - and take direct actions like soft delete, hard delete, or moving messages to junk or deleted folders. For manual remediation, Threat Explorer supports actions on emails delivered within the past 30 days In my next posts I will be delving further into how to use both the Graph APIs and the Security and Compliance PowerShell module to safely execute your purge commands.Search and Purge using the Security and Compliance PowerShell cmdlets
Welcome back to the series of blogs covering search and purge in Microsoft Purview eDiscovery! If you are new to this series, please first visit the blog post in our series that you can find here: Search and Purge workflow in the new modern eDiscovery experience. Also please ensure you read in full the Microsoft Learn documentation on this topic as I will not be covering some of the steps in full (permissions, releasing holds, all limitations): Find and delete email messages in eDiscovery | Microsoft Learn So as a reminder, E3/G3 customers must use the Security and Compliance PowerShell cmdlets to execute the purge operation. Searches can continue to be created using the New-ComplianceSearch cmdlet and then run the newly created search using the Start-ComplianceSearch cmdlet. Once a search has run, the statistics can be reviewed before executing the New-ComplianceSearchAction cmdlet with the Purge switch to remove the item from the targeted locations. However, some organizations may want to initially run the search, review statistics and export an item report in the new user experience before using the New-ComplianceSearchAction cmdlet to purge the items from the mailbox. Before starting, ensure you have version 3.9.0 or later of the Exchange Online Management PowerShell Module installed (link). If multiple versions of the Exchange Online Management PowerShell module are installed alongside version 3.9.0, remove the older versions of the module to avoid potential conflicts between the different versions of the module. When connecting using the Connect-IPPSession cmdlet ensure you include the EnableSearchOnlySession parameter otherwise the purge command will not run and may generate an error (link) Create the case, if you will be using the new Content Search case you can skip this step. However, if you want to create a new case to host the search, you must create the case via PowerShell. This ensures any searches created within the case in the Purview portal will support the PowerShell based purge command. Use the Connect-IPPSession command to connect to Security and Compliance PowerShell before running the following command to create a new case. New-ComplianceCase “Test Case” Select the new Purview Content Search case or the new case you created in step 1 and create a new Search Within your new search use the Add Sources option to search for and select the mailboxes containing the item to be purged by adding them to the Data sources of your newly created search. Note: Make sure only Exchange mailboxes are selected as you can only purge items contained within Exchange Mailboxes. If you added both the mailbox and associated sites, you can remove the sites using the 3 dot menu next to the data source under User Options. Alternatively, use the manage sources button to remove the sites associated with the data source. Within Condition builder define the conditions required to target the item you wish to purge. In this example, I am targeting an email with a specific subject, from a specific sender, on a specific day. To help me understand the estimated number of items that would be returned by the search I can run a statistics job first to give me confidence that the query is correct. I do this by selecting Run Query from the search itself. Then I can select Statistics and Run Query to trigger the Statistics job. Note, you can view the progress of the job via the Process Manager Once completed I can view the Statistics to confirm the query looks accurate and returning the numbers I was expecting. If I want to further verify that the items returned by the search is what I am looking for, I can run a Sample job to review a sample of the items matching the search query Once the Sample job is completed, I can review samples for locations with hits to determine if this is indeed the items I want to purge. If I need to go further and generate a report of the items that match the search (not just statistics and sampling) I can run an export to generate a report for the items that match the search criteria. Note: It is important to run the export report to review the results that purge action will remove from the mailbox. This will ensure that we purge only the items of interest. Download the report for the export job via the Process Manager or the Export tab to review the items that were a match Note: If very few locations have hits it is recommended to reduce the scope of your search by updating the data sources to include only the locations with hits. Switch back to the cmdlet and use Get-ComplianceSearch cmdlet as below, ensure the query is as you specified in the Purview Portal Get-ComplianceSearch -Identity "My search and purge" | fl As the search hasn’t be run yet in PowerShell – the Items count is 0 and the JobEndTime is not set - the search needs to be re-run via PS as per the example shown below Start-ComplianceSearch "My search and purge" Give it a few minutes to complete and use Get-ComplianceSearch to check the status of the search, if the status is not “Completed” and JobEndTime is not set you may need to give it more time Check the search returned the same results once it has finished running Get-ComplianceSearch -Identity "My search and purge" | fl name,status,searchtype,items,searchstatistics CRITICAL: It is important to make sure the Items count match the number of items returned in the item report generated from the Purview Portal. If the number of items returned in PowerShell do not match, then do not continue with the purge action. Issue the purge command using the New-ComplianceSearchAction cmdlet New-ComplianceSearchAction -SearchName "My search and purge" -Purge -PurgeType HardDelete Once completed check the status of the purge command to confirm that the items have been deleted Get-ComplianceSearchAction "My search and purge_purge" | fl Now that the purge operation has been completed successfully, it has been removed from the target mailbox and is no longer accessible by the user.
Enterprise Strategy for Secure Agentic AI: From Compliance to Implementation
Imagine an AI system that doesn’t just answer questions but takes action querying your databases, updating records, triggering workflows, even processing refunds without human intervention. That’s Agentic AI and it’s here. But with great power comes great responsibility. This autonomy introduces new attack surfaces and regulatory obligations. The Model Context Protocol (MCP) Server the gateway between your AI agent and critical systems becomes your Tier-0 control point. If it fails, the blast radius is enormous. This is the story of how enterprises can secure Agentic AI, stay compliant and implement Zero Trust architectures using Azure AI Foundry. Think of it as a roadmap a journey with three milestones - Milestone 1: Securing the Foundation Our journey starts with understanding the paradigm shift. Traditional AI with RAG (Retrieval-Augmented Generation) is like a librarian: It retrieves pre-indexed data. It summarizes information. It never changes the books or places orders. Security here is simple: protect the index, validate queries, prevent data leaks. But Agentic AI? It’s a staffer with system access. It can: Execute tools and business logic autonomously. Chain operations: read → analyze → write → notify. Modify data and trigger workflows. Bottom line: RAG is a “smart librarian.” Agentic AI is a “staffer with system access.” Treat the security model accordingly. And that means new risks: unauthorized access, privilege escalation, financial impact, data corruption. So what’s the defense? Ten critical security controls your first line of protection: Here’s what a production‑grade, Zero Trust MCP gateway needs. Its intentionally simplified in the demo (e.g., no auth) to highlight where you must harden in production. (https://github.com/davisanc/ai-foundry-mcp-gateway) Authentication Demo: None Prod: Microsoft Entra ID, JWT validation, Managed Identity, automatic credential rotation Authorization & RBAC Demo: None Prod: Tool‑level RBAC via Entra; least privilege; explicit allow‑lists per agent/capability Input Validation Demo: Basic (ext whitelist, 10MB, filename sanitize) Prod: JSON Schema validation, injection guards (SQL/command), business‑rule checks Rate Limiting Demo: None Prod: Multi‑tier (per‑agent, per‑tool, global), adaptive throttling, backoff Audit Logging Demo: Console → App Service logs Prod: Structured logs w/ correlation IDs, compliance metadata, PII redaction Session Management Demo: In‑memory UUID sessions Prod: Encrypted distributed storage (Redis/Cosmos DB), tenant isolation, expirations File Upload Security Demo: Ext whitelist, size limits, memory‑only Prod: 7‑layer defense (validate, MIME, malware scanning via Defender for Storage), encryption at rest, signed URLs Network Security Demo: Public App Service + HTTPS Prod: Private Endpoints, VNet integration, NSGs, Azure Firewall no public exposure Secrets Management Demo: App Service env vars (not in code) Prod: Azure Key Vault + Managed Identity, rotation, access audit Observability & Threat Detection (5‑Layer Stack) Layer 1: Application Insights (requests, dependencies, custom security events) Layer 2: Azure AI Content Safety (harmful content, jailbreaks) Layer 3: Microsoft Defender for AI (prompt injection incl. ASCII smuggling, credential theft, anomalous tool usage) Layer 4: Microsoft Purview for AI (PII/PHI classification, DLP on outputs, lineage, policy) Layer 5: Microsoft Sentinel (SIEM correlation, custom rules, automated response) Note: Azure AI Content Safety is built into Azure AI Foundry for real‑time filtering on both prompts and completions. Picture this as an airport security model: multiple checkpoints, each catching what the previous missed. That’s defense-in-depth. Zero Trust in Practice ~ A Day in the Life of a Prompt Every agent request passes through 8 sequential checkpoints, mapped to MITRE ATLAS tactics/mitigations (e.g., AML.M0011 Input Validation, AML.M0004 Output Filtering, AML.M0015 Adversarial Input Detection). The design goal is defense‑in‑depth: multiple independent controls, different detection signals, and layered failure modes. Checkpoints 1‑7: Enforcement (deny/contain before business systems) Checkpoint 8: Monitoring (detect/respond, hunt, learn, harden) AML.M0009 – Control Access to ML Models AML.M0011 – Validate ML Model Inputs AML.M0000 – Limit ML Model Availability AML.M0014 – ML Artifact Logging AML.M0004 – Output Filtering AML.M0015 – Adversarial Input Detection If one control slips, the others still stand. Resilience is the product of layers. Milestone 2: Navigating Compliance Next stop: regulatory readiness. The EU AI Act is the world’s first comprehensive AI law. If your AI system operates in or impacts the EU market, compliance isn’t optional, it’s mandatory. Agentic AI often falls under high-risk classification. That means: Risk management systems. Technical documentation. Logging and traceability. Transparency and human oversight. Fail to comply? Fines up to €30M or 6% of global turnover. Azure helps you meet these obligations: Entra ID for identity and RBAC. Purview for data classification and DLP. Defender for AI for prompt injection detection. Content Safety for harmful content filtering. Sentinel for SIEM correlation and incident response. And this isn’t just about today. Future regulations are coming US AI Executive Orders, UK AI Roadmap, ISO/IEC 42001 standards. The trend is clear: transparency, explainability, and continuous monitoring will be universal. Milestone 3: Implementation Deep-Dive Now, the hands-on part. How do you build this strategy into reality? Step 1: Entra ID Authentication Register your MCP app in Entra ID. Configure OAuth2 and JWT validation. Enable Managed Identity for downstream resources. Step 2: Apply the 10 Controls RBAC: Tool-level access checks. Validation: JSON schema + injection prevention. Rate Limiting: Express middleware or Azure API Management. Audit Logging: Structured logs with correlation IDs. Session Mgmt: Redis with encryption. File Security: MIME checks + Defender for Storage. Network: Private Endpoints + VNet. Secrets: Azure Key Vault. Observability: App Insights + Defender for AI + Purview + Sentinel. Step 3: Secure CI/CD Pipelines Embed compliance checks in Azure DevOps: Pre-build: Secret scanning. Build: RBAC & validation tests. Deploy: Managed Identity for service connections. Post-deploy: Compliance scans via Azure Policy. Step 4: Build the 5-Layer Observability Stack App Insights → Telemetry. Content Safety → Harmful content detection. Defender for AI → Prompt injection monitoring. Purview → PII/PHI classification and lineage. Sentinel → SIEM correlation and automated response. The Destination: A Secure, Compliant Future By now, you’ve seen the full roadmap: Secure the foundation with Zero Trust and layered controls. Navigate compliance with EU AI Act and prepare for global regulations. Implement the strategy using Azure-native tools and CI/CD best practices. Because in the world of Agentic AI, security isn’t optional, compliance isn’t negotiable, and observability is your lifeline. Resources https://learn.microsoft.com/en-us/azure/ai-foundry/what-is-azure-ai-foundry https://learn.microsoft.com/en-us/azure/defender-for-cloud/ai-threat-protection https://learn.microsoft.com/en-us/purview/ai-microsoft-purview https://atlas.mitre.org/ https://digital-strategy.ec.europa.eu/en/policies/european-approach-artificial-intelligence https://techcommunity.microsoft.com/blog/microsoft-security-blog/microsoft-sentinel-mcp-server---generally-available-with-exciting-new-capabiliti/4470125Empowering organizations with integrated data security: What’s new in Microsoft Purview
Today, data moves across clouds, apps, and devices at an unprecedented speed, often outside the visibility of siloed legacy tools. The rise of autonomous agents, generative AI, and distributed data ecosystems means that traditional perimeter-based security models are no longer sufficient. Even though companies are spending more than $213 billion globally, they still face several persistent security challenges: Fragmented tools don’t integrate together well and leave customers lacking full visibility of their data security risks The growing use of AI in the workplace is creating new data risks for companies to manage The shortage of skilled cybersecurity professionals is making it difficult to accomplish data security objectives Microsoft is a global leader in cloud, productivity, and security solutions. Microsoft Purview benefits from this breadth of offerings, integrating seamlessly across Microsoft 365, Azure, Microsoft Fabric, and other Microsoft platforms — while also working in harmony with complementary security tools. Unlike fragmented point solutions, Purview delivers an end-to-end data security platform built into the productivity and collaboration tools organizations already rely on. This deep understanding of data within Microsoft environments, combined with continually improving external data risk detections, allows customers to simplify their security stack, increase visibility, and act on data risks more quickly. At Ignite, we’re introducing the next generation of data security — delivering advanced protection and operational efficiency, so security teams can move at business speed while maintaining control of their data. Go beyond visibility into action, across your data estate Many customers today lack a comprehensive view of how to holistically address data security risks and properly manage their data security posture. To help customers strengthen data security across their data estate, we are excited to announce the new, enhanced Microsoft Purview Data Security Posture Management (DSPM). This new AI-powered DSPM experience unifies current Purview DSPM and DSPM for AI capabilities to create a central entry point for data security insights and controls, from which organizations can take action to continually improve their data security posture and prioritize risks. The new capabilities in the enhanced DSPM experience are: Outcome-Based workflows: Choose a data security objective and see related metrics, risk patterns, a recommended action plan and its impact - going from insight to action. Expanded coverage and remediation on Data Risk Assessments: Conduct item-level analysis with new remediation actions like bulk disabling of overshared SharePoint links. Out-of-box posture reports: Uncover data protection gaps and track security posture improvements with out-of-box reports that provide rich context on label usage, auto-labeling effectiveness, posture drift through label transitions, and DLP policy activities. AI Observability: Surface an organization’s agent inventory with assigned agent risk level and agent posture metrics based on agentic interactions with the organization’s data. New Security Copilot Agent: Accelerate the discovery and analysis of sensitive data to uncover hidden risks across files, emails, and messages. Gain visibility of non-Microsoft data within your data estate: Enable a unified view of data risks by gaining visibility into Salesforce, Snowflake, Google Cloud Platform, and Databricks – available through integrations with external partners via Microsoft Sentinel. These DSPM enhancements will be available in Public Preview within the upcoming weeks. Learn more in our blog dedicated to the announcement of the new Microsoft Purview DSPM. Together, these innovations reflect a larger shift: data security is no longer about silos—it’s about unified visibility and control everywhere data lives and having a comprehensive understanding of the data estate to detect and prevent data risks. Organizations trust Microsoft for their productivity and security platforms, but their footprint spans across third-party data environments too. That’s why Purview continues to expand protection beyond Microsoft environments. In addition to bringing in 3rd party data into DSPM, we are also expanding auto-labeling to three new Data Map sources, adding to the data sources we previously announced. Currently in public preview, the new sources include Snowflake, SQL Server, and Amazon S3. Once connected to Purview, admins gain an “at-a-glance” view of all data sources and can automatically apply sensitivity labels, enforcing consistent security policies without manual effort. This helps organizations discover sensitive information at scale, reduce the risk of data exposure, and ensure safer AI adoption all while simplifying governance through centralized policy management and visibility across their entire data estate. Enable AI adoption and prevent data oversharing As organizations adopt more autonomous agents, new risks emerge, such as unsupervised data access and creation, cascading agent interactions, and unclear data activity accountability. Besides AI Observability in DSPM providing details on the inventory and risk level of the agents, Purview is expanding its industry-leading data security and compliance capabilities to secure and govern agents that inherit users’ policies and controls, as well as agents that have their own unique IDs, policies, and controls. This includes agent types across Microsoft 365 Copilot, Copilot Studio, Microsoft Foundry, and third-party platforms. Key enhancements include: Extension of Purview Information Protection and Data Loss Prevention policies to autonomous agents: Scope autonomous agents with an Agent ID into Purview policies that work for users across Microsoft 365 apps, including Exchange, SharePoint, and Teams. Microsoft Purview Insider Risk Management for Agents: With dedicated indicators and behavioral analytics to flag specific risky agent activities, enable proactive investigation by assigning risk levels to each agent. Extension of Purview data compliance capabilities to agent interactions: Microsoft Purview Communication Compliance, Data Lifecycle Management, Audit, and eDiscovery extend to agent interactions, supporting responsible use, secure retention, and agentic accountability. Purview SDK embedded in Agent Framework SDK: Purview SDK embedded in Agent Framework SDK enables developers to integrate enterprise-grade security, compliance, and governance into AI agents. It delivers automatic data classification, prevents sensitive data leaks and oversharing, and provides visibility and control for regulatory compliance, empowering secure adoption of AI agents in complex environments. Purview integration with Foundry: Purview is now enabled within Foundry, allowing Foundry admins to activate Microsoft Purview on their subscription. Once enabled, interaction data from all apps and agents flows into Purview for centralized compliance, governance, and posture management of AI data. Azure AI Search honors Purview labels and policies: Azure AI Search now ingests Microsoft Purview sensitivity labels and enforces corresponding protection policies through built-in indexers (SharePoint, OneLake, Azure Blob, ADLS Gen2). This ensures secure, policy-aligned search over enterprise data, enabling agentic RAG scenarios where only authorized documents are returned or sent to LLMs, preventing oversharing and aligning with enterprise data protection standards. Extension of Purview Data Loss Prevention policies to Copilot Mode in Edge for Business: This week, Microsoft Edge for Business introduced Copilot Mode, transforming the browser into a proactive, agentic partner. This is AI-assisted browsing will honor the user’s existing DLP protections, such as endpoint DLP policies that prevent pasting to sensitive service domains, or summarizing sensitive page content. Learn more in our blog dedicated to the announcements of Microsoft Purview for Agents. New capabilities in Microsoft Purview, now in public preview, to help prevent data oversharing and leakage through AI include: Expansion of Microsoft Purview Data Loss Prevention (DLP) for Microsoft 365 Copilot: Previously, we introduced DLP for Microsoft 365 Copilot to prevent labeled files & emails from being used as grounding data for responses, therefore reducing the risk of oversharing. Today, we are expanding DLP for Microsoft 365 Copilot to safeguard prompts containing sensitive data. This real-time control helps organizations mitigate data leakage and oversharing risks by preventing Microsoft 365 Copilot, Copilot Chat, and Microsoft 365 Copilot agents from returning a response when prompts contain sensitive data or using that sensitive data for grounding in Microsoft 365 or the web. For example, if a user searches, “Can you tell me more about my customer based on their address: 1234 Main Street,” Copilot will both inform the user that organizational policies prevent it from responding to their prompt, as well as block any web queries to Bing for “1234 Main Street.” Enhancements to inline data protection in Edge for Business: Earlier this year, we introduced inline data protection in Edge for Business to prevent sensitive data from being leaked to unmanaged consumer AI apps, starting with ChatGPT, Google Gemini, and DeepSeek. We are not only making this capability generally available for the initial set of AI apps, but also expanding the capability to 30+ new apps in public preview and supporting file upload activity in addition to text. This addresses potential data leakage that can occur when employees send organizational files or data to consumer AI apps for help with work-related tasks, such as document creation or code reviews. Inline data protection for the network: For user activity outside of the browser, we are also enabling inline data protection at the network layer. Earlier this year, we introduced integrations with supported secure service edge (SSE) providers to detect when sensitive data is shared to unmanaged cloud locations, such as consumer AI apps or personal cloud storage, even if sharing occurs outside of the Edge browser. In addition to the discovery of sensitive data, these integrations now support protection controls that block sensitive data from leaving a user device and reaching an unmanaged cloud service or application. These capabilities are now generally available through the Netskope and iboss integrations, and inline data discovery is available in public preview through the Palo Alto Networks integration. Extension of Purview protection to on-device AI: Purview DLP policies now extend to the Recall experience in Copilot+ PC devices to prevent sensitive organizational data from being undesirably captured and retained. Admins can now block Recall snapshots based on sensitivity label or the presence of Purview sensitive information types (SITs) in a document open on the device, or simply honor and display the sensitivity labels of content captured in the Recall snapshot library. For example, a DLP policy can be configured to prevent recall from taking snapshots of any documents labeled “Highly Confidential,” or a product design file that contains intellectual property. Learn more in the Windows IT Pro blog. Best-in-class data security for Microsoft environments Microsoft Purview sets the standard for data security within its own ecosystem. Organizations benefit from unified security policies and seamless compliance controls that are purpose-built for Microsoft environments, ensuring sensitive data remains secure without compromising productivity. We also are constantly investing in expanding protections and controls to Microsoft collaboration tools including SharePoint, Teams, Fabric, Azure and across Microsoft 365. On-demand classification adds meeting transcript coverage and new enhancements: To help organizations protect sensitive data sitting in data-at-rest, on-demand classification now extends to meeting transcripts, enabling the discovery and classification of sensitive information shared in existing recorded meeting transcripts. Once classified, admins can set up DLP or Data Lifecycle Management (DLM) policies to properly protect and retain this data according to organizational policies. This is now generally available, empowering organizations to strengthen data security, streamline compliance, and ensure even sensitive information in data-at-rest is discovered, protected, and governed more effectively. In addition, on-demand classification for endpoints is also generally available, giving organizations even broader coverage across their data estate. New usage posture and consumption reports: We’re introducing new usage posture and consumption reports, now in public preview. Admins can quickly identify compliance gaps, optimize Purview seat assignments, and understand how consumptive features are driving spend. With granular insights by feature, policy, and user type, admins can analyze usage trends, forecast costs, and toggle consumptive features on and off directly, all from a unified dashboard. The result: stronger compliance, easier cost management, and better alignment of Purview investments to your organization’s needs. Enable DLP and Copilot protection with extended SharePoint permissions: Extended SharePoint permissions, now generally available, make it simple to protect and manage files in SharePoint by allowing library owners to apply a default sensitivity label to an entire document library. When this is enabled, the label is dynamically enforced across all unprotected files in the library, both new and existing, within the library. Downloaded files are automatically encrypted, and access is managed based on SharePoint site membership, giving organizations powerful, scalable access control. With extended SharePoint permissions, teams can consistently apply labels at scale, automate DLP policy enforcement, and confidently deploy Copilot, all without the need for manually labeling files. Whether for internal teams, external partners, or any group where permissions need to be tightly controlled, extended SharePoint permissions streamline protection and compliance in SharePoint. Network file filtering via Entra GSA integration: We are integrating Purview with Microsoft Entra to enable file filtering at the network layer. These filtering controls help prevent sensitive content from being shared to unauthorized services based on properties such as sensitivity labels or presence of Purview sensitive information types (SITs) within the file. For example, Entra admins can now create a file policy to block files containing credit card numbers from passing through the network. Learn more here. Expanded protection scenarios enabled by Purview endpoint DLP: We are introducing several noteworthy enhancements to Purview endpoint DLP to protect an even broader range of exfiltration or leakage scenarios from organizational devices, without hindering user productivity. These enhancements, initially available on Windows devices, include: Extending protection to unsaved files: Files no longer need to be saved to disk to be protected under a DLP policy. With this improvement, unsaved files will undergo a point-in-time evaluation to detect the presence of sensitive data and apply the appropriate protections. Expanded support for removable media: Admins can now prevent data exfiltration to broader list of removable media devices, including iPhones, Android devices, and CD-ROMs. Protection for Outlook attachments downloaded to removable media or network shares: Admins can now prevent exfiltration of email attachments when users attempt to drag and drop them into USB devices, network shares, and other removable media. Expanded capability support for macOS: In addition to the new endpoint DLP protections introduced above, we are also expanding the following capabilities, already available for Windows devices, to devices running on macOS: Expanded file type coverage to 110+ file types, blanket protections for non-Office or PDF file types, addition of “allow” and “off” policy actions, device-based policy scoping to scope policies to specific devices or device groups (or apply exclusions), and integration with Power Automate. Manageability and alert investigation improvements in Purview DLP: Lastly, we are also introducing device manageability and alert investigation improvements in Purview DLP to simplify the day-to-day experience for admins. These improvements include: Reporting and troubleshooting improvements for devices onboarded to endpoint DLP: We are introducing additional tools for admins to build confidence in their Purview DLP protections for endpoint devices. These enhancements, designed to maximize reliability and enable better troubleshooting of potential issues, include near real-time reporting of policy syncs initiated on devices and policy health insights into devices’ compliance status and readiness to receive policies. Enhancements to always-on diagnostics: Earlier this year, we introduced always-on diagnostics to automatically collect logs from Windows endpoint devices, eliminating the need to reproduce issues when submitting an investigation request or raising a support ticket. This capability is expanding so that admins now have on-demand access to diagnostic logs from users’ devices without intervening in their operations. This further streamlines the issue resolution process for DLP admins while minimizing end user disruption. Simplified DLP alert investigation, including easier navigation to crucial alert details in just 1 click, and the ability to aggregate alerts originating from a single user for more streamlined investigation and response. For organizations who manage Purview DLP alerts within their broader incident management process in Microsoft Defender, we are pleased to share that alert severities will now be synced between the Purview portal and the Defender portal. Expanding enterprise-grade data security to small and medium businesses (SMBs): Purview is extending its reach beyond large enterprises by introducing a new add-on for Microsoft 365 Business Premium, bringing advanced data security and compliance capabilities to SMBs. The Microsoft Purview suite for Business Premium brings the same enterprise-grade protection, such as sensitivity labeling, data loss prevention, and compliance management, to organizations with up to 300 users. This enables SMBs to operate with the same level of compliance and data security as large enterprises, all within a simplified, cost-effective experience built for smaller teams. Stepping into the new era of technology with AI-powered data security Globally, there is a shortage of skilled cybersecurity professionals. Simultaneously, the volume of alerts and incidents is ever growing. By infusing AI into data security solutions, admins can scale their impact. By reducing manual workloads, they enhance operational effectiveness and strengthen overall security posture – allowing defenders to stay ahead. In 2025, 82% of organizations have developed plans to use GenAI to fortify their data security programs. With its cutting-edge generative AI-powered investigative capabilities, Microsoft Purview Data Security Investigations (DSI) is transforming and scaling how data security admins analyze incident-related data. Since being released into public preview in April, the product has made a big impact with customers like Toyota Motors North America. "Data Security Investigations eliminates manual work, automating investigations in minutes. It’s designed to handle the scale and complexity of large data sets by correlating user activity with data movement, giving analysts a faster, more efficient path to meaningful insights,” said solution architect Dan Garawecki. This Ignite, we are introducing several new capabilities in DSI, including: DSI integration with DSPM: View proactive, summary insights and launch a Data Security Investigation directly from DSPM. This integration brings the full power of DSI analysis to your fingertips, enabling admins to drill into data risks surfaced in DSPM with speed and precision. Enhancements in DSI AI-powered deep content analysis capabilities: Admins can now add context before AI analysis for higher-quality, more efficient investigations. A new AI-powered natural language search function lets admins locate specific files using keywords, metadata, and embeddings. Vector search and content categorization enhancements allow admins to better identify risky assets. Together, these enhancements equip admins with sharper, faster tools for identifying buried data risks – both proactively and reactively. DSI cost transparency report and in-product estimator: To help customers manage pay-as-you-go billing, DSI is adding a new lightweight in-product cost estimator and transparency report. We are also expanding Security Copilot in Microsoft Purview with AI-powered capabilities that strengthen both the protection and investigation of sensitive data by introducing the Data Security Posture Agent and Data Security Triage Agent. Data Security Posture Agent: Available in preview, the new Data Security Posture Agent uses LLMs to help admins answer “Is this happening?” across thousands of files—delivering fast, intent-driven discovery and risk profiling, even when explicit keywords are absent. Integrated with Purview DSPM, it surfaces actionable insights and improves compliance, helping teams reduce risk and respond to threats before they escalate. Data Security Triage Agent: Alongside this, the Data Security Triage Agent, now generally available, enables analysts to efficiently triage and remediate the most critical alerts, automating incident response and surfacing the threats that matter most. Together, these agentic capabilities convert high-volume signals into consistent, closed-loop action, accelerate investigations and remediation, reduce policy-violation dwell time, and improve audit readiness, all natively integrated within Microsoft 365 and Purview so security teams can scale outcomes without scaling headcount. To make the agents easily accessible and help teams get started more quickly, we are excited to announce that Security Copilot will be available to all Microsoft 365 E5 customers. Rollout starts today for existing Security Copilot customers with Microsoft 365 E5 and will continue in the upcoming months for all Microsoft 365 E5 customers. Customers will receive advanced notice before activation. Learn more: https://aka.ms/SCP-Ignite25 Data security that keeps innovating alongside you As we look ahead, Microsoft Purview remains focused on empowering organizations with scalable solutions that address the evolving challenges of data security. While we deliver best-in-class security for Microsoft, we recognize that today’s organizations rarely operate in a single cloud, many businesses rely on a diverse mix of platforms to power their operations and innovation. That’s why we have been extending Purview’s capabilities beyond Microsoft environments, helping customers protect data across their entire digital estate. In a world where data is the lifeblood of innovation, securing it must be more than a checkbox—it must be a catalyst for progress. As organizations embrace AI, autonomous agents, and increasingly complex digital ecosystems, Microsoft Purview empowers them to move forward with confidence. By unifying visibility, governance, and protection across the entire data estate, Purview transforms security from a fragmented challenge into a strategic advantage. The future of data security isn’t just about defense—it’s about enabling bold, responsible innovation at scale. Let’s build that future together.Uncover hidden security risks with Microsoft Sentinel graph
Earlier this fall, we launched Microsoft Sentinel graph – and today, we are pleased to announce that Sentinel graph is generally available starting December 1, 2025. Microsoft Sentinel graph maps the interconnections across activity, asset, and threat intelligence data. This enables comprehensive graph-based security and analysis across pre-and post-breach scenarios in both Microsoft Defender and Microsoft Purview. Customers are already seeing the impact of the graph-powered experiences that is providing insights beyond tabular queries. "The predefined scenarios in Sentinel graph are excellent... it definitely shows where I would need to look as an investigator to figure out what's happening in my environment, who has access to it, not only directly, but also indirectly, a couple of hops away. And that's something that you really can't get through a standard KQL query..." - Gary Bushey, Security Architect, Cyclotron, Inc. Building on this foundation, we are taking Sentinel graph to the next level and are excited to announce the public preview of the following new capabilities. Graph MCP Tools Building on the hunting graph and blast radius analysis capabilities in Microsoft Defender portal. We are excited to announce preview of purpose-built Sentinel graph MCP tools (Blast Radius, Path Discovery, and Exposure Perimeter) that make the graph-powered insights accessible to the AI agents. Using these purpose-built Sentinel graph MCP tools, you will be able to use and build AI agents to get insights from the graph in natural language (figure 1): “What is the blast radius from ‘Laura Hanak’?” “Is there a path from user Mark Gafarov to key vault wg-prod?” “Who can all get to wg-prod key vault?” You can sign up here for a free preview of Sentinel graph MCP tools, which will also roll out starting December 1, 2025. Custom Graphs The security operations teams, including Tier-3 analysts, threat intelligence specialists, and security researchers play a critical role in investigating sophisticated attacks and addressing systemic security issues. Their responsibilities range from uncovering design vulnerabilities and tracing historical exploitation, to analyzing types of abuse and recommending effective solutions. These experts strive to identify hidden patterns within organizational data and struggle with the right tools that can help them differentiate between normal vs. abnormal, keep-up with the changing attack patterns, and handle massive and complex datasets at scale. This requires a high level of flexibility and customization to rapidly iterate on the analysis. We’re taking Microsoft Sentinel graph to the next level and are thrilled to announce the public preview of custom graphs with two new powerful approaches designed specifically for security: ephemeral custom graphs and materialized custom graphs. These innovative approaches empower defenders to create and analyze graphs tailored and tuned to their unique security scenarios to find hidden risks and patterns in their security data available in the Sentinel data lake. Using their data in the lake, defenders will be able author notebooks (figure 2) to model, build, visualize, traverse, and run advanced graph analyses like Chokepoint/Centrality, Blast Radius/Reachability, Prioritized Path/Ranked, and K-hop. It’s a transformative leap in graph analytics, fundamentally changing how security teams understand and mitigate organizational risk by connecting the dots in their data. Figure 2: Custom graphs using Notebook in VS Code You can sign up here for a free preview of custom graph capability, which will also roll out starting December 1, 2025. Ephemeral Custom Graphs Ephemeral custom graphs are for one-time investigations requiring quick pattern examination and rapidly changing large scale data that doesn't justify materialization for reuse. For example, in a typical SOC investigation, brute-force attempts or privilege escalations appear as isolated incidents. But in reality, attackers move laterally through interconnected credentials and resources. Let’s assume, a service account (svc-backup) used by a legacy database is compromised. It holds group membership in “DataOps-Admins,” which shares access with “Engineering-All.” A developer reuses their personal access token across staging and production clusters. Individually, these facts seem harmless. Together, they form a multi-hop credential exposure chain that can only be detected through graph traversal. Sentinel graph helps you to build ad-hoc graphs for an investigation and discarded afterward (not kept in a database for reuse). You can pull the data from the Sentinel data lake and build a graph to explore relationships, run analytics, iterate on nodes/edges, and refine queries in an interactive loop. Here are some additional scenarios where ephemeral custom graphs can expose hidden patterns: Sign-in anomaly hunting: An analyst graphs user logins against source IPs and timestamps to identify unusual patterns (like a single IP connecting to many accounts). By iterating on the graph (filtering nodes, adding context like geolocation), they can spot suspicious login clusters or a credential theft scenario. TTP (Tactics, Techniques, Procedures) investigation: For a specific threat (e.g., a known APT’s techniques), the hunter might use a graph template to map related events. Microsoft Sentinel, for instance, can provide hunting notebook templates for scenarios like investigating lateral movement or scanning logs for leaked credentials, so analysts quickly construct a graph of relevant evidence. Audit log pattern discovery: By graphing Office 365 activity logs or admin audit logs, defenders can apply advanced graph algorithms (like betweenness centrality) to find outliers – e.g., an account that intermediates many rare files access relationships might indicate insider abuse. Materialized Custom Graphs Materialized custom graphs are graph datasets that are stored and maintained over time, often updated at intervals (e.g., daily or hourly). Instead of being thrown away each session, these graphs will be materialized in the graph database for running graph analytics and visualization. Materialized custom graphs will enable organizations to create their custom enterprise knowledge graphs for various use cases, such as every organization already has an identity graph — they just haven’t visualized it yet. Imagine a large enterprise where users, devices, service principals, and applications are constantly changing. New credentials are issued, groups evolve, and permissions shift by the hour. Over time, this churn creates a complex web of implicit trust and shared access that no static tool can capture. Organizations can now build their own identity graphs and materialize them. These materialized custom graphs can continuously map relationships across Azure AD Domain Services, Entra ID, AWS IAM, SaaS platforms, and custom applications, updating daily or hourly to reflect the organization’s true security topology. Organizations can query these graphs and run various advanced graph algorithms and understand the chokepoint, blast radius, attack paths, and so on. This helps detect the gradual buildup of privilege overlap — when identities that were once isolated begin to share access paths through evolving group memberships, role assignments, or inherited permissions. Over weeks or months, these subtle shifts expand the blast radius of any single compromise. Behind the scenes We are partnering with our friends in Microsoft Fabric to bring these new capabilities to market. Mapping a large digital estate into a graph requires new scale out approach and that is what graph in Microsoft Fabric enables. “Discovering modern security risks is a massive data challenge. It requires connecting the dots across an entire digital estate, which can only be achieved with a graph at hyperscale. This is why our Fabric team's partnership with the Sentinel graph team is so critical. We’ve collaborated to build a scale-out graph solution capable of processing billion nodes and edges, delivering the performance and scale our largest security customers need to stay ahead of threats.” - Yitzhak Kesselman, CVP, Fabric Real-Time Intelligence Getting started Check out this video to learn more. To get access to the preview capabilities, please sign-up here. Reference links Data lake blog MCP server blogSecurity as the core primitive in the agentic era: New innovations to secure AI agents and apps
This week at Microsoft Ignite, we shared our vision for Microsoft security -- In the agentic era, security must be ambient and autonomous, like the AI it protects. It must be woven into and around everything we build—from silicon to OS, to agents, apps, data, platforms, and clouds—and throughout everything we do. In this blog, we are going to dive deeper into many of the new innovations we are introducing this week to secure AI agents and apps. As I spend time with our customers and partners, there are four consistent themes that have emerged as core security challenges to secure AI workloads. These are: preventing agent sprawl and access to resources, protecting against data oversharing and data leaks, defending against new AI threats and vulnerabilities, and adhering to evolving regulations. Addressing these challenges holistically requires a coordinated effort across IT, developers, and security leaders, not just within security teams and to enable this, we are introducing several new innovations: Microsoft Agent 365 for IT, Foundry Control Plane in Microsoft Foundry for developers, and the Security Dashboard for AI for security leaders. In addition, we are releasing several new purpose-built capabilities to protect and govern AI apps and agents across Microsoft Defender, Microsoft Entra, and Microsoft Purview. Observability at every layer of the stack To facilitate the organization-wide effort that it takes to secure and govern AI agents and apps – IT, developers, and security leaders need observability (security, management, and monitoring) at every level. IT teams need to enable the development and deployment of any agent in their environment. To ensure the responsible and secure deployment of agents into an organization, IT needs a unified agent registry, the ability to assign an identity to every agent, manage the agent’s access to data and resources, and manage the agent’s entire lifecycle. In addition, IT needs to be able to assign access to common productivity and collaboration tools, such as email and file storage, and be able to observe their entire agent estate for risks such as over-permissioned agents. Development teams need to build and test agents, apply security and compliance controls by default, and ensure AI models are evaluated for safety guardrails and security vulnerabilities. Post deployment, development teams must observe agents to ensure they are staying on task, accessing applications and data sources appropriately, and operating within their cost and performance expectations. Security & compliance teams must ensure overall security of their AI estate, including their AI infrastructure, platforms, data, apps, and agents. They need comprehensive visibility into all their security risks- including agent sprawl and resource access, data oversharing and leaks, AI threats and vulnerabilities, and complying with global regulations. They want to address these risks by extending their existing security investments that they are already invested in and familiar with, rather than using siloed or bolt-on tools. These teams can be most effective in delivering trustworthy AI to their organizations if security is natively integrated into the tools and platforms that they use every day, and if those tools and platforms share consistent security primitives such as agent identities from Entra; data security and compliance controls from Purview; and security posture, detections, and protections from Defender. With the new capabilities being released today, we are delivering observability at every layer of the AI stack, meeting IT, developers, and security teams where they are in the tools they already use to innovate with confidence. For IT Teams - Introducing Microsoft Agent 365, the control plane for agents, now in preview The best infrastructure for managing your agents is the one you already use to manage your users. With Agent 365, organizations can extend familiar tools and policies to confidently deploy and secure agents, without reinventing the wheel. By using the same trusted Microsoft 365 infrastructure, productivity apps, and protections, organizations can now apply consistent and familiar governance and security controls that are purpose-built to protect against agent-specific threats and risks. gement and governance of agents across organizations Microsoft Agent 365 delivers a unified agent Registry, Access Control, Visualization, Interoperability, and Security capabilities for your organization. These capabilities work together to help organizations manage agents and drive business value. The Registry powered by the Entra provides a complete and unified inventory of all the agents deployed and used in your organization including both Microsoft and third-party agents. Access Control allows you to limit the access privileges of your agents to only the resources that they need and protect their access to resources in real time. Visualization gives organizations the ability to see what matters most and gain insights through a unified dashboard, advanced analytics, and role-based reporting. Interop allows agents to access organizational data through Work IQ for added context, and to integrate with Microsoft 365 apps such as Outlook, Word, and Excel so they can create and collaborate alongside users. Security enables the proactive detection of vulnerabilities and misconfigurations, protects against common attacks such as prompt injections, prevents agents from processing or leaking sensitive data, and gives organizations the ability to audit agent interactions, assess compliance readiness and policy violations, and recommend controls for evolving regulatory requirements. Microsoft Agent 365 also includes the Agent 365 SDK, part of Microsoft Agent Framework, which empowers developers and ISVs to build agents on their own AI stack. The SDK enables agents to automatically inherit Microsoft's security and governance protections, such as identity controls, data security policies, and compliance capabilities, without the need for custom integration. For more details on Agent 365, read the blog here. For Developers - Introducing Microsoft Foundry Control Plane to observe, secure and manage agents, now in preview Developers are moving fast to bring agents into production, but operating them at scale introduces new challenges and responsibilities. Agents can access tools, take actions, and make decisions in real time, which means development teams must ensure that every agent behaves safely, securely, and consistently. Today, developers need to work across multiple disparate tools to get a holistic picture of the cybersecurity and safety risks that their agents may have. Once they understand the risk, they then need a unified and simplified way to monitor and manage their entire agent fleet and apply controls and guardrails as needed. Microsoft Foundry provides a unified platform for developers to build, evaluate and deploy AI apps and agents in a responsible way. Today we are excited to announce that Foundry Control Plane is available in preview. This enables developers to observe, secure, and manage their agent fleets with built-in security, and centralized governance controls. With this unified approach, developers can now identify risks and correlate disparate signals across their models, agents, and tools; enforce consistent policies and quality gates; and continuously monitor task adherence and runtime risks. Foundry Control Plane is deeply integrated with Microsoft’s security portfolio to provide a ‘secure by design’ foundation for developers. With Microsoft Entra, developers can ensure an agent identity (Agent ID) and access controls are built into every agent, mitigating the risk of unmanaged agents and over permissioned resources. With Microsoft Defender built in, developers gain contextualized alerts and posture recommendations for agents directly within the Foundry Control Plane. This integration proactively prevents configuration and access risks, while also defending agents from runtime threats in real time. Microsoft Purview’s native integration into Foundry Control Plane makes it easy to enable data security and compliance for every Foundry-built application or agent. This allows Purview to discover data security and compliance risks and apply policies to prevent user prompts and AI responses from safety and policy violations. In addition, agent interactions can be logged and searched for compliance and legal audits. This integration of the shared security capabilities, including identity and access, data security and compliance, and threat protection and posture ensures that security is not an afterthought; it’s embedded at every stage of the agent lifecycle, enabling you to start secure and stay secure. For more details, read the blog. For Security Teams - Introducing Security Dashboard for AI - unified risk visibility for CISOs and AI risk leaders, coming soon AI proliferation in the enterprise, combined with the emergence of AI governance committees and evolving AI regulations, leaves CISOs and AI risk leaders needing a clear view of their AI risks, such as data leaks, model vulnerabilities, misconfigurations, and unethical agent actions across their entire AI estate, spanning AI platforms, apps, and agents. 90% of security professionals, including CISOs, report that their responsibilities have expanded to include data governance and AI oversight within the past year. 1 At the same time, 86% of risk managers say disconnected data and systems lead to duplicated efforts and gaps in risk coverage. 2 To address these needs, we are excited to introduce the Security Dashboard for AI. This serves as a unified dashboard that aggregates posture and real-time risk signals from Microsoft Defender, Microsoft Entra, and Microsoft Purview. This unified dashboard allows CISOs and AI risk leaders to discover agents and AI apps, track AI posture and drift, and correlate risk signals to investigate and act across their entire AI ecosystem. For example, you can see your full AI inventory and get visibility into a quarantined agent, flagged for high data risk due to oversharing sensitive information in Purview. The dashboard then correlates that signal with identity insights from Entra and threat protection alerts from Defender to provide a complete picture of exposure. From there, you can delegate tasks to the appropriate teams to enforce policies and remediate issues quickly. With the Security Dashboard for AI, CISOs and risk leaders gain a clear, consolidated view of AI risks across agents, apps, and platforms—eliminating fragmented visibility, disconnected posture insights, and governance gaps as AI adoption scales. Best of all, there’s nothing new to buy. If you’re already using Microsoft security products to secure AI, you’re already a Security Dashboard for AI customer. Figure 5: Security Dashboard for AI provides CISOs and AI risk leaders with a unified view of their AI risk by bringing together their AI inventory, AI risk, and security recommendations to strengthen overall posture Together, these innovations deliver observability and security across IT, development, and security teams, powered by Microsoft’s shared security capabilities. With Microsoft Agent 365, IT teams can manage and secure agents alongside users. Foundry Control Plane gives developers unified governance and lifecycle controls for agent fleets. Security Dashboard for AI provides CISOs and AI risk leaders with a consolidated view of AI risks across platforms, apps, and agents. Added innovation to secure and govern your AI workloads In addition to the IT, developer, and security leader-focused innovations outlined above, we continue to accelerate our pace of innovation in Microsoft Entra, Microsoft Purview, and Microsoft Defender to address the most pressing needs for securing and governing your AI workloads. These needs are: Manage agent sprawl and resource access e.g. managing agent identity, access to resources, and permissions lifecycle at scale Prevent data oversharing and leaks e.g. protecting sensitive information shared in prompts, responses, and agent interactions Defend against shadow AI, new threats, and vulnerabilities e.g. managing unsanctioned applications, preventing prompt injection attacks, and detecting AI supply chain vulnerabilities Enable AI governance for regulatory compliance e.g. ensuring AI development, operations, and usage comply with evolving global regulations and frameworks Manage agent sprawl and resource access 76% of business leaders expect employees to manage agents within the next 2–3 years. 3 Widespread adoption of agents is driving the need for visibility and control, which includes the need for a unified registry, agent identities, lifecycle governance, and secure access to resources. Today, Microsoft Entra provides robust identity protection and secure access for applications and users. However, organizations lack a unified way to manage, govern, and protect agents in the same way they manage their users. Organizations need a purpose-built identity and access framework for agents. Introducing Microsoft Entra Agent ID, now in preview Microsoft Entra Agent ID offers enterprise-grade capabilities that enable organizations to prevent agent sprawl and protect agent identities and their access to resources. These new purpose-built capabilities enable organizations to: Register and manage agents: Get a complete inventory of the agent fleet and ensure all new agents are created with an identity built-in and are automatically protected by organization policies to accelerate adoption. Govern agent identities and lifecycle: Keep the agent fleet under control with lifecycle management and IT-defined guardrails for both agents and people who create and manage them. Protect agent access to resources: Reduce risk of breaches, block risky agents, and prevent agent access to malicious resources with conditional access and traffic inspection. Agents built in Microsoft Copilot Studio, Microsoft Foundry, and Security Copilot get an Entra Agent ID built-in at creation. Developers can also adopt Entra Agent ID for agents they build through Microsoft Agent Framework, Microsoft Agent 365 SDK, or Microsoft Entra Agent ID SDK. Read the Microsoft Entra blog to learn more. Prevent data oversharing and leaks Data security is more complex than ever. Information Security Media Group (ISMG) reports that 80% of leaders cite leakage of sensitive data as their top concern. 4 In addition to data security and compliance risks of generative AI (GenAI) apps, agents introduces new data risks such as unsupervised data access, highlighting the need to protect all types of corporate data, whether it is accessed by employees or agents. To mitigate these risks, we are introducing new Microsoft Purview data security and compliance capabilities for Microsoft 365 Copilot and for agents and AI apps built with Copilot Studio and Microsoft Foundry, providing unified protection, visibility, and control for users, AI Apps, and Agents. New Microsoft Purview controls safeguard Microsoft 365 Copilot with real-time protection and bulk remediation of oversharing risks Microsoft Purview and Microsoft 365 Copilot deliver a fully integrated solution for protecting sensitive data in AI workflows. Based on ongoing customer feedback, we’re introducing new capabilities to deliver real-time protection for sensitive data in M365 Copilot and accelerated remediation of oversharing risks: Data risk assessments: Previously, admins could monitor oversharing risks such as SharePoint sites with unprotected sensitive data. Now, they can perform item-level investigations and bulk remediation for overshared files in SharePoint and OneDrive to quickly reduce oversharing exposure. Data Loss Prevention (DLP) for M365 Copilot: DLP previously excluded files with sensitivity labels from Copilot processing. Now in preview, DLP also prevents prompts that include sensitive data from being processed in M365 Copilot, Copilot Chat, and Copilot agents, and prevents Copilot from using sensitive data in prompts for web grounding. Priority cleanup for M365 Copilot assets: Many organizations have org-wide policies to retain or delete data. Priority cleanup, now generally available, lets admins delete assets that are frequently processed by Copilot, such as meeting transcripts and recordings, on an independent schedule from the org-wide policies while maintaining regulatory compliance. On-demand classification for meeting transcripts: Purview can now detect sensitive information in meeting transcripts on-demand. This enables data security admins to apply DLP policies and enforce Priority cleanup based on the sensitive information detected. & bulk remediation Read the full Data Security blog to learn more. Introducing new Microsoft Purview data security capabilities for agents and apps built with Copilot Studio and Microsoft Foundry, now in preview Microsoft Purview now extends the same data security and compliance for users and Copilots to agents and apps. These new capabilities are: Enhanced Data Security Posture Management: A centralized DSPM dashboard that provides observability, risk assessment, and guided remediation across users, AI apps, and agents. Insider Risk Management (IRM) for Agents: Uniquely designed for agents, using dedicated behavioral analytics, Purview dynamically assigns risk levels to agents based on their risky handing of sensitive data and enables admins to apply conditional policies based on that risk level. Sensitive data protection with Azure AI Search: Azure AI Search enables fast, AI-driven retrieval across large document collections, essential for building AI Apps. When apps or agents use Azure AI Search to index or retrieve data, Purview sensitivity labels are preserved in the search index, ensuring that any sensitive information remains protected under the organization’s data security & compliance policies. For more information on preventing data oversharing and data leaks - Learn how Purview protects and governs agents in the Data Security and Compliance for Agents blog. Defend against shadow AI, new threats, and vulnerabilities AI workloads are subject to new AI-specific threats like prompt injections attacks, model poisoning, and data exfiltration of AI generated content. Although security admins and SOC analysts have similar tasks when securing agents, the attack methods and surfaces differ significantly. To help customers defend against these novel attacks, we are introducing new capabilities in Microsoft Defender that deliver end-to-end protection, from security posture management to runtime defense. Introducing Security Posture Management for agents, now in preview As organizations adopt AI agents to automate critical workflows, they become high-value targets and potential points of compromise, creating a critical need to ensure agents are hardened, compliant, and resilient by preventing misconfigurations and safeguarding against adversarial manipulation. Security Posture Management for agents in Microsoft Defender now provides an agent inventory for security teams across Microsoft Foundry and Copilot Studio agents. Here, analysts can assess the overall security posture of an agent, easily implement security recommendations, and identify vulnerabilities such as misconfigurations and excessive permissions, all aligned to the MITRE ATT&CK framework. Additionally, the new agent attack path analysis visualizes how an agent’s weak security posture can create broader organizational risk, so you can quickly limit exposure and prevent lateral movement. Introducing Threat Protection for agents, now in preview Attack techniques and attack surfaces for agents are fundamentally different from other assets in your environment. That’s why Defender is delivering purpose-built protections and detections to help defend against them. Defender is introducing runtime protection for Copilot Studio agents that automatically block prompt injection attacks in real time. In addition, we are announcing agent-specific threat detections for Copilot Studio and Microsoft Foundry agents coming soon. Defender automatically correlates these alerts with Microsoft’s industry-leading threat intelligence and cross-domain security signals to deliver richer, contextualized alerts and security incident views for the SOC analyst. Defender’s risk and threat signals are natively integrated into the new Microsoft Foundry Control Plane, giving development teams full observability and the ability to act directly from within their familiar environment. Finally, security analysts will be able to hunt across all agent telemetry in the Advanced Hunting experience in Defender, and the new Agent 365 SDK extends Defender’s visibility and hunting capabilities to third-party agents, starting with Genspark and Kasisto, giving security teams even more coverage across their AI landscape. To learn more about how you can harden the security posture of your agents and defend against threats, read the Microsoft Defender blog. Enable AI governance for regulatory compliance Global AI regulations like the EU AI Act and NIST AI RMF are evolving rapidly; yet, according to ISMG, 55% of leaders report lacking clarity on current and future AI regulatory requirements. 5 As enterprises adopt AI, they must ensure that their AI innovation aligns with global regulations and standards to avoid costly compliance gaps. Introducing new Microsoft Purview Compliance Manager capabilities to stay ahead of evolving AI regulations, now in preview Today, Purview Compliance Manager provides over 300 pre-built assessments for common industry, regional, and global standards and regulations. However, the pace of change for new AI regulations requires controls to be continuously re-evaluated and updated so that organizations can adapt to ongoing changes in regulations and stay compliant. To address this need, Compliance Manager now includes AI-powered regulatory templates. AI-powered regulatory templates enable real-time ingestion and analysis of global regulatory documents, allowing compliance teams to quickly adapt to changes as they happen. As regulations evolve, the updated regulatory documents can be uploaded to Compliance Manager, and the new requirements are automatically mapped to applicable recommended actions to implement controls across Microsoft Defender, Microsoft Entra, Microsoft Purview, Microsoft 365, and Microsoft Foundry. Automated actions by Compliance Manager further streamline governance, reduce manual workload, and strengthen regulatory accountability. Introducing expanded Microsoft Purview compliance capabilities for agents and AI apps now in preview Microsoft Purview now extends its compliance capabilities across agent-generated interactions, ensuring responsible use and regulatory alignment as AI becomes deeply embedded across business processes. New capabilities include expanded coverage for: Audit: Surface agent interactions, lifecycle events, and data usage with Purview Audit. Unified audit logs across user and agent activities, paired with traceability for every agent using an Entra Agent ID, support investigation, anomaly detection, and regulatory reporting. Communication Compliance: Detect prompts sent to agents and agent-generated responses containing inappropriate, unethical, or risky language, including attempts to manipulate agents into bypassing policies, generating risky content, or producing noncompliant outputs. When issues arise, data security admins get full context, including the prompt, the agent’s output, and relevant metadata, so they can investigate and take corrective action Data Lifecycle Management: Apply retention and deletion policies to agent-generated content and communication flows to automate lifecycle controls and reduce regulatory risk. Read about Microsoft Purview data security for agents to learn more. Finally, we are extending our data security, threat protection, and identity access capabilities to third-party apps and agents via the network. Advancing Microsoft Entra Internet Access Secure Web + AI Gateway - extend runtime protections to the network, now in preview Microsoft Entra Internet Access, part of the Microsoft Entra Suite, has new capabilities to secure access to and usage of GenAI at the network level, marking a transition from Secure Web Gateway to Secure Web and AI Gateway. Enterprises can accelerate GenAI adoption while maintaining compliance and reducing risk, empowering employees to experiment with new AI tools safely. The new capabilities include: Prompt injection protection which blocks malicious prompts in real time by extending Azure AI Prompt Shields to the network layer. Network file filtering which extends Microsoft Purview to inspect files in transit and prevents regulated or confidential data from being uploaded to unsanctioned AI services. Shadow AI Detection that provides visibility into unsanctioned AI applications through Cloud Application Analytics and Defender for Cloud Apps risk scoring, empowering security teams to monitor usage trends, apply Conditional Access, or block high-risk apps instantly. Unsanctioned MCP server blocking prevents access to MCP servers from unauthorized agents. With these controls, you can accelerate GenAI adoption while maintaining compliance and reducing risk, so employees can experiment with new AI tools safely. Read the Microsoft Entra blog to learn more. As AI transforms the enterprise, security must evolve to meet new challenges—spanning agent sprawl, data protection, emerging threats, and regulatory compliance. Our approach is to empower IT, developers, and security leaders with purpose-built innovations like Agent 365, Foundry Control Plane, and the Security Dashboard for AI. These solutions bring observability, governance, and protection to every layer of the AI stack, leveraging familiar tools and integrated controls across Microsoft Defender, Microsoft Entra, and Microsoft Purview. The future of security is ambient, autonomous, and deeply woven into the fabric of how we build, deploy, and govern AI systems. Explore additional resources Learn more about Security for AI solutions on our webpage Learn more about Microsoft Agent 365 Learn more about Microsoft Entra Agent ID Get started with Microsoft 365 Copilot Get started with Microsoft Copilot Studio Get started with Microsoft Foundry Get started with Microsoft Defender for Cloud Get started with Microsoft Entra Get started with Microsoft Purview Get started with Microsoft Purview Compliance Manager Sign up for a free Microsoft 365 E5 Security Trial and Microsoft Purview Trial 1 Bedrock Security, 2025 Data Security Confidence Index, published Mar 17, 2025. 2 AuditBoard & Ascend2, Connected Risk Report 2024; as cited by MIT Sloan Management Review, Spring 2025. 3 KPMG AI Quarterly Pulse Survey | Q3 2025. September 2025. n= 130 U.S.-based C-suite and business leaders representing organizations with annual revenue of $1 billion or more 4 First Annual Generative AI study: Business Rewards vs. Security Risks, , Q3 2023, ISMG, N=400 5 First Annual Generative AI study: Business Rewards vs. Security Risks, Q3 2023, ISMG, N=400Microsoft Security Store: Now Generally Available
When we launched the Microsoft Security Store in public preview on September 30, our goal was simple: make it easier for organizations to discover, purchase, and deploy trusted security solutions and AI agents that integrate seamlessly with Microsoft Security products. Today, Microsoft Security Store is generally available—with three major enhancements: Embedded where you work: Security Store is now built into Microsoft Defender, featuring SOC-focused agents, and into Microsoft Entra for Verified ID and External ID scenarios like fraud protection. By bringing these capabilities into familiar workflows, organizations can combine Microsoft and partner innovation to strengthen security operations and outcomes. Expanded catalog: Security Store now offers more than 100 third-party solutions, including advanced fraud prevention, forensic analysis, and threat intelligence agents. Security services available: Partners can now list and sell services such as managed detection and response and threat hunting directly through Security Store. Real-World Impact: What We Learned in Public Preview Thousands of customers explored Microsoft Security Store and tried a growing catalog of agents and SaaS solutions. While we are at the beginning of our journey, customer feedback shows these solutions are helping teams apply AI to improve security operations and reduce manual effort. Spairliners, a cloud-first aviation services joint venture between Air France and Lufthansa, strengthened identity and access controls by deploying Glueckkanja’s Privileged Admin Watchdog to enforce just-in-time access. “Using the Security Store felt easy, like adding an app in Entra. For a small team, being able to find and deploy security innovations in minutes is huge.” – Jonathan Mayer, Head of Innovation, Data and Quality GTD, a Chilean technology and telecommunications company, is testing a variety of agents from the Security Store: “As any security team, we’re always looking for ways to automate and simplify our operations. We are exploring and applying the world of agents more and more each day so having the Security Store is convenient—it’s easy to find and deploy agents. We’re excited about the possibilities for further automation and integrations into our workflows, like event-triggered agents, deeper Outlook integration, and more." – Jonathan Lopez Saez, Cybersecurity Architect Partners echoed the momentum they are seeing with the Security Store: “We’re excited by the early momentum with Security Store. We’ve already received multiple new leads since going live, including one in a new market for us, and we have multiple large deals we’re looking to drive through Security Store this quarter.” - Kim Brault, Head of Alliances, Delinea “Partnering with Microsoft through the Security Store has unlocked new ways to reach enterprise customers at scale. The store is pivotal as the industry shifts toward AI, enabling us to monetize agents without building our own billing infrastructure. With the new embedded experience, our solutions appear at the exact moment customers are looking to solve real problems. And by working with Microsoft’s vetting process, we help provide customers confidence to adopt AI agents” – Milan Patel, Co-founder and CEO, BlueVoyant “Agents and the Microsoft Security Store represent a major step forward in bringing AI into security operations. We’ve turned years of service experience into agentic automations, and it’s resonating with customers—we’ve been positively surprised by how quickly they’re adopting these solutions and embedding our automated agentic expertise into their workflows.” – Christian Kanja, Founder and CEO of glueckkanja New at GA: Embedded in Defender, Entra—Security Solutions right where you work Microsoft Security Store is now embedded in the Defender and Entra portals with partner solutions that extend your Microsoft Security products. By placing Security Store in front of security practitioners, it’s now easier than ever to use the best of partner and Microsoft capabilities in combination to drive stronger security outcomes. As Dorothy Li, Corporate Vice President of Security Copilot and Ecosystem put it, “Embedding the Security Store in our core security products is about giving customers access to innovative solutions that tap into the expertise of our partners. These solutions integrate with Microsoft Security products to complete end-to-end workflows, helping customers improve their security” Within the Microsoft Defender portal, SOC teams can now discover Copilot agents from both Microsoft and partners in the embedded Security Store, and run them all from a single, familiar interface. Let’s look at an example of how these agents might help in the day of the life of a SOC analyst. The day starts with Watchtower (BlueVoyant) confirming Sentinel connectors and Defender sensors are healthy, so investigations begin with full visibility. As alerts arrive, the Microsoft Defender Copilot Alert Triage Agent groups related signals, extracts key evidence, and proposes next steps; identity related cases are then validated with Login Investigator (adaQuest), which baselines recent sign-in behavior and device posture to cut false positives. To stay ahead of emerging campaigns, the analyst checks the Microsoft Threat Intelligence Briefing Agent for concise threat rundowns tied to relevant indicators, informing hunts and temporary hardening. When HR flags an offboarding, GuardianIQ (People Tech Group) correlates activity across Entra ID, email, and files to surface possible data exfiltration with evidence and risk scores. After containment, Automated Closing Comment Generator (Ascent Global Inc.) produces clear, consistent closure notes from Defender incident details, keeping documentation tight without hours of writing. Together, these Microsoft and partner agents maintain platform health, accelerate triage, sharpen identity decisions, add timely threat context, reduce insider risk blind spots, and standardize reporting—all inside the Defender portal. You can read more about the new agents available in the Defender portal in this blog. In addition, Security Store is now integrated into Microsoft Entra, focused on identity-centric solutions. Identity admins can discover and activate partner offerings for DDoS protection, intelligent bot defense, and government ID–based verification for account recovery —all within the Entra portal. With these capabilities, Microsoft Entra delivers a seamless, multi-layered defense that combines built-in identity protection with best-in-class partner technologies, making it easier than ever for enterprises to strengthen resilience against modern identity threats. Learn more here. Levent Besik, VP of Microsoft Entra, shared that “This sets a new benchmark for identity security and partner innovation at Microsoft. Attacks on digital identities can come from anywhere. True security comes from defense in depth, layering protection across the entire user journey so every interaction, from the first request to identity recovery, stays secure. This launch marks only the beginning; we will continue to introduce additional layers of protection to safeguard every aspect of the identity journey” New at GA: Services Added to a Growing Catalog of Agents and SaaS For the first time, partners can offer their security services directly through the Security Store. Customers can now find, buy, and activate managed detection and response, threat hunting, and other expert services—making it easier to augment internal teams and scale security operations. Every listing has a MXDR Verification that certifies they are providing next generation advanced threat detection and response services. You can browse all the services available at launch here, and read about some of our exciting partners below: Avanade is proud to be a launch partner for professional services in the Microsoft Security Store. As a leading global Microsoft Security Services provider, we’re excited to make our offerings easier to find and help clients strengthen cyber defenses faster through this streamlined platform - Jason Revill, Avanade Global Security Technology Lead ProServeIT partnering with Microsoft to have our offers in the Microsoft Security Store helps ProServeIT protect our joint customers and allows us to sell better with Microsoft sellers. It shows customers how our technology and services support each other to create a safe and secure platform - Eric Sugar, President Having Reply’s security services showcased in the Microsoft Security Store is a significant milestone for us. It amplifies our ability to reach customers at the exact point where they evaluate and activate Microsoft security solutions, ensuring our offerings are visible alongside Microsoft’s trusted technologies. Notable New Selections Since public preview, the Security Store catalog has grown significantly. Customers can now choose from over 100 third-party solutions, including 60+ SaaS offerings and 50+ Security Copilot agents, with new additions every week. Recent highlights include Cisco Duo and Rubrik: Cisco Duo IAM delivers comprehensive, AI-driven identity protection combining MFA, SSO, passwordless and unified directory management. Duo IAM seamlessly integrates across the Microsoft Security suite—enhancing Entra ID with risk-based authentication and unified access policy management across cloud and on-premises applications seamlessly in just a few clicks. Intune for device compliance and access enforcement. Sentinel for centralized security monitoring and threat detection through critical log ingestion about authentication events, administrator actions, and risk-based alerts, providing real-time visibility across the identity stack. Rubrik's data security platform delivers complete cyber resilience across enterprise, cloud, and SaaS alongside Microsoft. Through the Microsoft Sentinel integration, Rubrik’s data management capabilities are combined with Sentinel’s security analytics to accelerate issue resolution, enabling unified visibility and streamlined responses. Furthermore, Rubrik empowers organizations to reduce identity risk and ensure operational continuity with real-time protection, unified visibility and rapid recovery across Microsoft Active Directory and Entra ID infrastructure. The Road Ahead This is just the beginning. Microsoft Security Store will continue to make it even easier for customers to improve their security outcomes by tapping into the innovation and expertise of our growing partner ecosystem. The momentum we’re seeing is clear—customers are already gaining real efficiencies and stronger outcomes by adopting AI-powered agents. As we work together with partners, we’ll unlock even more automation, deeper integrations, and new capabilities that help security teams move faster and respond smarter. Explore the Security Store today to see what’s possible. For a more detailed walk-through of the capabilities, read our previous public preview Tech Community post If you’re a partner, now is the time to list your solutions and join us in shaping the future of security.
Announcing New Microsoft Purview Capabilities to Protect GenAI Agents
As organizations accelerate their adoption of agentic AI, a new and urgent challenge is emerging: how to protect the rapidly growing number of agents—first-party, third-party, and custom-built—created and deployed across the enterprise. At Microsoft Ignite, we’re introducing major advancements in Microsoft Purview to support our customers in securing all their agents, wherever they operate. The New Reality: Agents Everywhere, Data Risks Amplified For many organizations, the pace of agent creation is outstripping traditional oversight. Developers, business units, and other information workers can spin up agents to automate tasks, analyze data, or interact with enterprise systems. This proliferation brings tremendous opportunity, but also a new level of risk. New agents can access sensitive information, trigger cascading actions of other agents, and operate outside direct human supervision. The anxiety is real: how do you protect every agent, even those you didn’t know existed? Data risks are especially critical in this new landscape. Agents can process and share sensitive information at scale, interact with external systems, and invoke other agents or large language models, multiplying the complexity and potential of data exposure. Unlike traditional apps, agents are dynamic, autonomous, and currently often invisible to standard security controls. The risk surface expands with every new agent, making comprehensive data protection not just a technical requirement, but a business imperative. Purview for Agent 365: Protections for a more complex agent world Therefore, this week, we are announcing Agent 365 (A365) as the control plane for agents, enabling organizations to confidently manage, secure, and govern AI agents—first-party, third-party, and custom-built—across the enterprise. With A365, teams can extend familiar Microsoft 365 tools and policies to agents, ensuring unified inventory, robust access controls, lifecycle management, and deep integration with productivity apps. That’s why we’re extending Microsoft Purview protections to A365, bringing enterprise-grade security, compliance, and risk management to every agent. Here’s what we’re introducing to make this possible: AI Observability in Data Security Posture Management: Organizations gain visibility, risk assessment, and guided remediation for agents across Microsoft environments. Note: While third-party agents are included in the inventory, assigned risk levels, risk patterns, and guided remediation currently apply to M365 Copilot agents, Copilot Studio, and Microsoft Foundry agents. AI Observability in Data Security Posture Management Agentic Risk in Insider Risk Management: New agent-specific risk indicators and behavioral analytics enable precise detection and policy enforcement. For example, organizations can now identify risky agent behaviors, such as unauthorized data access or unusual activity patterns, and take targeted action to mitigate threats. Data Loss Prevention (DLP) and Information Protection controls extended to agent actions: Purview DLP and Information Protection policies now extend to agents that operate autonomously, allowing these agents to inherit the same protections and organizational policies as users. For example, these built-in controls ensure AI agents don’t access or share sensitive data when accessing M365 data within apps, whether that means blocking access for agents to labeled files or preventing agents from sending external emails and Teams messages that contain sensitive data. Expanded governance via Communication Compliance, Audit, Data Lifecycle Management and eDiscovery: Organizations benefit from expanded proactive detection, secure retention, and policy-based governance for interactions between humans and agents. By including these protections in A365, organizations can apply Purview’s enterprise-grade security, compliance, and risk controls to every agent—making it simpler and safer for customers to deploy agents at scale. Learn more about the Agent 365 announcement. Extending Purview Controls for All Agents Not all agents in an organization will run under an A365 license, yet every agent still requires strong data security and compliance controls. For that reason, we are also adding the following Purview capabilities: Inline data protection for prompts and responses: Expanded DLP capabilities prevent the sharing of sensitive data or files between users and third-party agents (such as ChatGPT or Google Gemini in Agent mode) through inline data protection for the browser and network. Purview SDK embedded in Agent Framework SDK: Purview SDK embedded in Agent Framework SDK enables developers to seamlessly integrate enterprise-grade security, compliance, and governance into the AI agents they build. This integration enables automatic classification and protection of sensitive data, prevents data leaks and oversharing, and provides visibility and control for regulatory compliance—empowering organizations to confidently and securely adopt AI agents in complex environments. Embedding Security into the Foundry Development Pipeline We are also adding several Purview capabilities specifically available in Foundry: Purview integration with Foundry: Purview is now enabled within Foundry, allowing Foundry admins to activate Microsoft Purview on their subscription. Once enabled, interaction data from all apps and agents flows into Purview for centralized compliance, governance, and posture management of AI data. Azure AI Search honors Purview labels and policies: Azure AI Search now ingests Microsoft Purview sensitivity labels and enforces corresponding protection policies through built-in indexers (SharePoint, OneLake, Azure Blob, ADLS Gen2). This enables secure, policy-aligned search over enterprise data, enabling agentic RAG scenarios where only authorized documents are returned or sent to LLMs, preventing data oversharing and aligning with enterprise data protection standards. Communication Compliance for Foundry: New policies extend Communication Compliance capabilities to Foundry, allowing security admins to set organization-wide Communication Compliance policies for acceptable communication for interactions with Foundry-built apps and agents, supported by Microsoft’s Responsible AI Standard. In Foundry Control Plane, Foundry admins will be able to view any deviations from this policy. In addition, Purview admins will be able to review potentially risky AI interactions in Communication Compliance, enabling them to decide on appropriate next steps. Communication Compliance provides visibility to potential unethical or harmful agent interactions Automated AI Compliance Assessments: The new integration between Microsoft Purview Compliance Manager and Foundry delivers automated, real-time compliance for AI solutions. Organizations can quickly assess agents against global standards like the EU AI Act, NIST AI RMF, and ISO/IEC, with one-click assessments and live monitoring of critical controls such as fairness, safety, and transparency. This streamlined approach eliminates manual mapping, surfaces actionable insights, and helps AI systems remain audit-ready as they evolve. Strengthening Trust in Microsoft 365 Copilot And we’re not stopping there. We’re continuing to expand Purview’s protections for Microsoft 365 Copilot to help organizations provide real-time protection for sensitive data in and accelerate remediation of oversharing risks. New enhancements include: Item-level oversharing investigation and remediation: Data security admins can now use data risk assessments in DSPM to analyze user sharing links in SharePoint and OneDrive and take bulk actions such as applying sensitivity labels to shared files, requesting the site owner to review sharing links, or disabling the links entirely. These enhancements streamline risk management, reduce exposure, and give organizations greater control over sensitive data at scale. Expanding DLP for Microsoft 365 Copilot to safeguard sensitive prompts and prevent data leakage: This new real-time control applicable to M365 Copilot, Copilot Chat and Copilot agents, helps prevent data leaks and oversharing by detecting and blocking sensitive data based on SITs in prompts. By blocking the prompt, this also prevents sensitive data from being used for grounding in Microsoft 365 or the web. This expands on the existing capability to prevent sensitive files & emails from being accessed by Copilot based on sensitivity label. Data security and compliance admins need stronger controls for Copilot-related assets like Teams meeting recordings and transcripts. They want to identify recordings with sensitive data and delete them to reduce risk and free up storage. We are announcing two new capabilities to help: Priority cleanup for M365 Copilot assets: Enables admins to override existing retention policies and compliantly delete files, such as meetings recordings and transcripts created to support Copilot use. Priority cleanup is now generally available in Purview Data Lifecycle Management. On-demand classification now extends to meeting transcripts: Information Protection automatically classifies files when they’re created, accessed, or modified, identifying sensitive information in real time. On-demand classification brings the same discovery and classification to data-at-rest without requiring user interactions. We’ve now added meeting transcripts to that coverage. Once the sensitive data in meeting transcripts is discovered and classified, admins can apply DLP or Data Lifecycle Management (DLM) to protect sensitive data from being shared or exposed unintentionally. Honoring Purview data security controls when using Copilot Mode in Edge for Business: Microsoft Edge for Business now features Copilot Mode to empower users to accelerate their productivity through AI-assisted browsing. Copilot Mode honors existing Purview data protections, such as preventing summarization of sensitive content open in the browser. Additionally, Agent Mode can be enabled for multi-step agent workflows in the browser. These agentic workflows will honor the user’s existing DLP protections, such as endpoint DLP policies that prevent pasting of sensitive data to sensitive service domains. Collectively, these capabilities reinforce Purview as the enterprise standard for securing AI-powered productivity. They give organizations the protections they need to scale Copilot usage with confidence and control. Empowering Secure Agentic AI Adoption As agents become integral to enterprise operations, Purview’s expanded protections empower organizations to safely embrace agentic AI—maintaining control, trust, and accountability at every step. With unified data security and compliance, organizations can observe and assess agent risk, prevent oversharing and leakage, detect risky agent behavior, and take decisive control to turn agentic AI into a trusted engine for growth. To learn more about Agent 365, visit the Agent 365 website.
