Recent Blogs
Containers are the backbone of modern cloud-native apps — and increasingly, the infrastructure powering AI, from AI assistants to a new wave of intelligent agents. They also blur the line between bui...
Jun 16, 2026
As organizations accelerate adoption of AI agents across Microsoft 365 and enterprise environments, security teams face a fundamental shift:
Agents are becoming first-class identities and securing ...
Jun 16, 2026
Co-authored with Lizet Pena, Caroline Mutua, Alvin Kua and Marco Sudahl
Security operations teams today are being asked to do more than ever: respond faster, manage increasing data volumes, reduce ...
Jun 16, 2026
Today, Microsoft announces the general availability of Microsoft’s Signing Transparency (MST) – a first-of-its-kind capability that brings unprecedented visibility and trust to our software supply ch...
Jun 15, 2026Recent Discussions
Ways to fetch quarantine files
We are working with quarantine files and have a few questions: 1. Is there a public API available to retrieve quarantined files from Microsoft Defender for Endpoint? 2. Is there a documented method to map an alert or a file SHA-1/SHA-256 hash to the corresponding object in the Defender quarantine store? 3. Is there a way to retrieve quarantined files other than using a PowerShell script through the Live Response API?
Co Authoring with Sensitivity Labels
Hello, I am working with sensitivity labels with my organization. We currently have Standard, Confidential, and Highly Confidential which all are encrypted. I have Co-Authoring turned on but I have some trouble with. We a lot of documents being collaborated on. Standard: Co-Authoring functions normal and Auto-Save is toggled on. Highly Confidential: Custom Permission in Sensitivity Label (View, Edit, Reply, Forward) I asked copilot and it stated even though my permissions are selected custom I have "Edit" on their for my internal users it is reading it as Co authoring; Co-Authoring is on and functioning but internal end users Auto-Save is toggled off and they are being asked to save a copy of the document or excel sheet then upload it again to SharePoint. Why isn't "Auto-Save" toggled on for "Highly Confidential" label? Can it be adjusted so it can be on? Do I have to make adjustments to my permissions in the Sensitivity label? Any help is appreciated. Thank you!Windows 11 24H2 Sec Baseline → Broken SSO to on‑prem (Root cause: PKINIT SHA‑1 baseline)
Hi all, I ran into an issue with Entra-joined devices using Windows Hello for Business (Cloud Kerberos Trust) that might help others working with Windows 11 24H2 security baselines. Scenario Windows 11 25H2 devices Entra-joined (not hybrid) Intune-managed Windows Hello for Business (WHfB) enabled Cloud Kerberos Trust configured On-prem AD (Windows Server 2019/2022 DCs) Access to SMB shares / on-prem applications Symptoms SSO to on-prem resources fails Users get credential/PIN prompt instead of SSO Error message: “The system cannot contact a domain controller to service the authentication request” Client-side observations: klist → no tickets (initially) After enabling Cloud Kerberos Trust: klist get krbtgt → works klist get cifs/server.domain → fails Error: 0xc000a100 / 0x3bc4 Hash generation for the specified version and hash type is not enabled on server Root Cause The issue was caused by a Windows 11 24H2 security baseline setting related to Kerberos/PKINIT. The 24H2 baseline introduces a policy for configuring hash algorithms for certificate-based Kerberos authentication (PKINIT). This setting allows environments to disable SHA-1 and require SHA-2 algorithms. [applepie.se] Important detail: This configuration only works if the domain controllers fully support PKINIT with SHA-2, which effectively requires Windows Server 2025 domain controllers across the environment. If SHA-1 is disabled while running: Windows Server 2019 or 2022 DCs Mixed environments then PKINIT authentication fails, which directly impacts: Windows Hello for Business Cloud Kerberos Trust Any passwordless Kerberos-based authentication Why this is difficult to troubleshoot Cloud Kerberos Trust appears correctly configured AzureADKerberos object exists PRT is valid Network connectivity is fine However: Kerberos tickets are not issued correctly Service tickets (CIFS, HTTP, etc.) fail Errors are misleading and point to KDC/hash issues No explicit warning is provided in baseline guidance that mixed environments will break Resolution Revert the baseline change and allow SHA-1 for PKINIT again. Policy location: Computer Configuration → System → Kerberos / KDC → Configure hash algorithms for certificate logon Ensure: SHA-1 is set to Allowed/Default After reverting: Kerberos ticket issuance works SSO to on-prem resources is restored Recommendation Do not disable SHA-1 for PKINIT unless: All domain controllers are Windows Server 2025, and PKINIT SHA-2 support has been fully validated Treat this setting as future hardening, not production-safe for mixed environments today. Takeaway If you experience: WHfB + Cloud Kerberos Trust SSO failures klist get errors with hash generation issues Missing or failing Kerberos service tickets check the PKINIT hash configuration from the 24H2 security baseline first.
Anthropic Claude Purview Data Connector showing all users as Guests..
It appears this connector is not mapping fields properly causing internal users to be mapped as "guests", and since prompts/data isn't maintained for guest users the connector is effectively not gathering anything but noise. Unlike the other data connectors, one cannot create field mappings. Also the app being named using the guid of Microsoft's own "dataassessments" service principal I don't think is intended either. Has anybody else experienced this? See below for an example.
Microsoft Defender for Endpoint and WDAC audit logs not include kernel audit/blocks
While testing WDAC on a fully patched Win11 pro machine - I noticed that kernel audit/block events do not get collected by MDE in the advanced hunting portal, only user mode audit/blocks are collected. Can anyone confirm they see this too and is this by design? My test case is to use a Strict Kernel Mode WDAC policy (as per: https://github.com/HotCakeX/Harden-Windows-Security/wiki/WDAC-policy-for-BYOVD-Kernel-mode-only-protection) which is active, using the global secure access client as my test, when the machine boots, the below event is generated locally on the machine: This event is never shown on the MDE advanced hunting portal, though user events do show. Examples of events that are coming through: Not receiving these events centrally for auditing would make deploying a kernel mode wdac control impossible. Would be amazing if Microsoft product team could look into this and resolve as these alerts should be captured as well please to facilitate deployment of more secure controls.
Microsoft 365 Developer E5 license lacking endpoints and device on defender portal
Dear Support Team, I am a microsoft certified trainer (MCT). I currently have a Microsoft 365 Developer E5 license assigned to my tenant. However, I have noticed that my Microsoft Defender portal (security.microsoft.com) is missing several critical features. For example, I cannot see the Endpoints or Devices menus, which is preventing me from implementing and testing Microsoft Defender for Endpoint. Additionally, my Azure tenant and Microsoft 365 tenant are separate. This has created challenges when configuring security services such as Microsoft Sentinel (SIEM), as certain prerequisites and integrations require configuration through the Microsoft Defender portal. Due to the missing Defender features, I am unable to complete the necessary setup. I would appreciate your assistance in understanding: Why the Endpoints and Devices sections are unavailable in my Defender portal despite having a Microsoft 365 Developer E5 license. Whether additional licensing, onboarding steps, or tenant configurations are required to enable Microsoft Defender for Endpoint features. How best to integrate or align my separate Azure and Microsoft 365 tenants to support services such as Microsoft Sentinel and Defender XDR. These issues are significantly impacting my ability to evaluate and implement Microsoft's security solutions. I would appreciate any guidance or recommendations to resolve them. Thank you for your assistance. Kind regards, [Your Name]Prompted to sign in to Microsoft Defender Platform on W11/W2025 using Entra
Hi Microsoft Defender XDR community, Since around May 18th, our users on devices that are onboarded to Microsoft Defender for Endpoint are being prompted to sign-in to the following application using Entra on login to Windows. Application Microsoft Defender Platform Application ID cab96880-db5b-4e15-90a7-f3f1d62ffe39 Is anyone aware of a change that requires user sign-in to Entra as a requirement for Microsoft Defender for Endpoint? I have tried raising a support topic on this topic. Regards ChrisCrowdstrike Agent with Microsoft Purview DLP
Could we use CrowdStrike as the main (Active) EDR while also enrolling the same machine into the agent-based Purview DLP? We have currently deployed MDE (passive) through the MDE Portal Onboarding, with RTP (Real-Time Protection) and BM (Behavioral Monitoring) enabled in the policy settings. While testing policies against user devices, we are unable to generate any alerts that match a rule based on conditions (e.g. PII, CC Data - where a user tries to copy and print sensitive information in a document), and the action based on the rule should be to BLOCK. This is not happening because there seems to be a disconnect from the workstation receiving the policies from Purview.Campaign-Centric Hunting with Microsoft Defender XDR and Microsoft Sentinel
Phishing investigations usually start with one suspicious email. A user reports a message. An alert is generated. An analyst opens the email details, checks the sender, reviews the URL, and tries to understand whether the message is malicious. That is a normal starting point. However, in a real SOC investigation, one email is rarely the full story. Attackers usually operate in campaigns. They reuse sender infrastructure, similar subjects, URLs, payloads, templates, and delivery techniques. A single email may be only one part of a wider phishing or malware campaign targeting multiple users. This is why campaign-centric hunting is important. I wrote this article from the perspective of a SOC analyst who often needs to move quickly from a single suspicious email to the full campaign impact. The goal is simple: use Microsoft Defender XDR and Microsoft Sentinel together to understand who was targeted, what was delivered, who clicked, and what should be prioritized first. Why Campaign-Centric Hunting When investigating a phishing or malware email, analysts usually need to answer practical questions: How many users received messages from the same campaign? Were the messages blocked, junked, delivered, or remediated? Did any user click the URL? Did anyone click through a Safe Links warning? Were any priority or high-risk users affected? Was the email removed after delivery? Are there related Defender XDR or Sentinel incidents? If we only investigate one message, we may miss the bigger picture. Campaign-centric hunting helps the SOC move from this question: Is this email malicious? To this question: What is the full impact of this campaign? That shift is important because the response priority should be based on campaign impact, not only on a single alert. What Campaign Views Provides Campaign Views in Microsoft Defender for Office 365 help analysts investigate coordinated email attacks such as phishing and malware campaigns. From Campaign Views, analysts can review campaign-level information such as: Campaign name Campaign type Campaign subtype Targeted users Inboxed messages Clicked users Visited links Sender domains Sender IPs Payload URLs Delivery actions Campaign timeline Campaign flow This is useful during triage because it quickly shows whether an email is part of a wider attack. For example, one reported phishing message may look small at first. But if Campaign Views shows that the same campaign targeted 50 users, delivered messages to 15 inboxes, and had 2 users click the URL, the investigation becomes much more urgent. Where CampaignInfo Fits The CampaignInfo table gives analysts a KQL-based way to query campaign-related data. Some useful fields are: Field Purpose CampaignId Unique identifier for the campaign CampaignName Name of the campaign CampaignType Campaign category, such as Phish or Malware CampaignSubtype Additional context, such as brand being phished or malware family NetworkMessageId Unique identifier for the email message RecipientEmailAddress Recipient affected by the campaign Timestamp Time when the event was recorded For correlation, the most important field is usually: NetworkMessageId This field can help connect campaign data with other Defender XDR email tables, including: EmailEvents UrlClickEvents EmailPostDeliveryEvents EmailAttachmentInfo EmailUrlInfo This makes CampaignInfo a useful pivot table for campaign-level hunting. Important note: CampaignInfo is currently documented as Preview. Before using these queries in production analytics rules, validate the table availability, schema, and results in your own tenant. Practical Scenario An analyst receives a phishing alert in Microsoft Defender XDR. The alert is related to a user who received a suspicious email with a credential-harvesting URL. The analyst opens Campaign Views and sees that the message belongs to a wider phishing campaign. At that point, the investigation should not stop with the original user. The analyst should now ask: Who else received this campaign? How many messages were delivered? Which users clicked? Did any users click through the Safe Links warning? Were the messages removed after delivery? Are there related incidents in Microsoft Sentinel? The investigation flow could look like this: Start from Campaign Views in Microsoft Defender XDR. Identify the campaign details. Use CampaignInfo to list affected users and messages. Join with EmailEvents to validate delivery status. Join with UrlClickEvents to identify user interaction. Join with EmailPostDeliveryEvents to confirm remediation. Review related Microsoft XDR incidents in Microsoft Sentinel. Prioritize response based on campaign impact. Query 1: List Recent Campaigns The first query gives a simple overview of recent campaigns. CampaignInfo | where Timestamp > ago(14d) | summarize FirstSeen = min(Timestamp), LastSeen = max(Timestamp), AffectedUsers = dcount(RecipientEmailAddress), Messages = dcount(NetworkMessageId) by CampaignId, CampaignName, CampaignType, CampaignSubtype | order by LastSeen desc This helps analysts quickly identify campaigns that affected the organization during the selected period. Useful questions to ask from this output: Which campaigns are most recent? Which campaigns affected the most users? Are the campaigns phishing, malware, or spam? Is there a specific brand or malware family in the subtype? Are similar campaigns appearing repeatedly? Query 2: Understand Delivery Impact After identifying campaigns, the next step is to understand delivery impact. A campaign that was fully blocked is different from a campaign that reached user inboxes. let Campaigns = CampaignInfo | where Timestamp > ago(14d) | project CampaignId, CampaignName, CampaignType, CampaignSubtype, NetworkMessageId, RecipientEmailAddress; Campaigns | join kind=leftouter ( EmailEvents | where Timestamp > ago(14d) | project NetworkMessageId, RecipientEmailAddress, Subject, SenderFromAddress, SenderFromDomain, SenderIPv4, DeliveryAction, DeliveryLocation, ThreatTypes, DetectionMethods, Timestamp ) on NetworkMessageId, RecipientEmailAddress | summarize Messages = dcount(NetworkMessageId), AffectedUsers = dcount(RecipientEmailAddress), Subjects = make_set(Subject, 5), SenderDomains = make_set(SenderFromDomain, 10), SenderIPs = make_set(SenderIPv4, 10) by CampaignId, CampaignName, CampaignType, CampaignSubtype, DeliveryAction, DeliveryLocation | order by AffectedUsers desc, Messages desc This query helps separate campaigns that were blocked from campaigns that actually reached users. From a SOC perspective, delivered messages deserve closer attention, especially if they reached the inbox. Query 3: Identify Users Who Clicked Campaign URLs Delivery is important, but clicks usually increase the priority of the incident. This query joins campaign data with UrlClickEvents. let Campaigns = CampaignInfo | where Timestamp > ago(14d) | project CampaignId, CampaignName, CampaignType, CampaignSubtype, NetworkMessageId, RecipientEmailAddress; Campaigns | join kind=inner ( UrlClickEvents | where Timestamp > ago(14d) | project NetworkMessageId, AccountUpn, Url, ActionType, IsClickedThrough, ThreatTypes, DetectionMethods, IPAddress, Workload, ClickTime = Timestamp ) on NetworkMessageId | summarize FirstClick = min(ClickTime), LastClick = max(ClickTime), ClickEvents = count(), ClickedUsers = dcount(AccountUpn), ClickThroughUsers = dcountif(AccountUpn, IsClickedThrough == true), ClickedUrls = make_set(Url, 10), SourceIPs = make_set(IPAddress, 10) by CampaignId, CampaignName, CampaignType, CampaignSubtype | order by ClickThroughUsers desc, ClickedUsers desc, LastClick desc This query helps identify campaigns where users interacted with the payload. If a user clicked a phishing URL, the next step should usually include identity-focused investigation, such as reviewing sign-in activity, MFA status, session activity, and possible risky sign-ins. Query 4: Focus on Click-Through Events Safe Links may block access to a malicious site. In some cases, however, a user may continue through a warning page. Those cases should be reviewed carefully. let Campaigns = CampaignInfo | where Timestamp > ago(30d) | project CampaignId, CampaignName, CampaignType, CampaignSubtype, NetworkMessageId, RecipientEmailAddress; Campaigns | join kind=inner ( UrlClickEvents | where Timestamp > ago(30d) | where IsClickedThrough == true | project NetworkMessageId, AccountUpn, Url, ActionType, ThreatTypes, IPAddress, ClickTime = Timestamp ) on NetworkMessageId | project ClickTime, CampaignId, CampaignName, CampaignType, CampaignSubtype, AccountUpn, RecipientEmailAddress, Url, ActionType, ThreatTypes, IPAddress | order by ClickTime desc This is one of the most useful views during incident response. A click-through event does not automatically mean compromise, but it is a strong reason to investigate the user account further. Query 5: Confirm Post-Delivery Remediation A malicious message may be delivered first and removed later by ZAP, AIR, or manual remediation. This query joins CampaignInfo with EmailPostDeliveryEvents. let Campaigns = CampaignInfo | where Timestamp > ago(30d) | project CampaignId, CampaignName, CampaignType, CampaignSubtype, NetworkMessageId, RecipientEmailAddress; Campaigns | join kind=leftouter ( EmailPostDeliveryEvents | where Timestamp > ago(30d) | project NetworkMessageId, RecipientEmailAddress, RemediationTime = Timestamp, Action, ActionType, ActionTrigger, ActionResult, DeliveryLocation, SourceLocation ) on NetworkMessageId, RecipientEmailAddress | summarize RemediatedMessages = dcountif(NetworkMessageId, isnotempty(ActionType)), RemediationTypes = make_set(ActionType, 10), RemediationResults = make_set(ActionResult, 10), LastRemediation = max(RemediationTime) by CampaignId, CampaignName, CampaignType, CampaignSubtype | order by LastRemediation desc This helps answer a very important question: Were the delivered malicious messages actually removed? This is useful for both SOC triage and reporting because it shows not only detection, but also response. Query 6: Campaign Blast Radius Summary The following query combines campaign, delivery, click, and remediation data into one campaign-level view. let TimeRange = 30d; let Campaigns = CampaignInfo | where Timestamp > ago(TimeRange) | project CampaignId, CampaignName, CampaignType, CampaignSubtype, NetworkMessageId, RecipientEmailAddress; let Delivery = EmailEvents | where Timestamp > ago(TimeRange) | summarize DeliveryActions = make_set(DeliveryAction, 10), DeliveryLocations = make_set(DeliveryLocation, 10), DeliveredMessages = dcountif(NetworkMessageId, DeliveryAction =~ "Delivered"), JunkedMessages = dcountif(NetworkMessageId, DeliveryAction =~ "Junked"), BlockedMessages = dcountif(NetworkMessageId, DeliveryAction =~ "Blocked"), Subjects = make_set(Subject, 5), SenderDomains = make_set(SenderFromDomain, 10) by NetworkMessageId, RecipientEmailAddress; let Clicks = UrlClickEvents | where Timestamp > ago(TimeRange) | summarize ClickEvents = count(), ClickThroughEvents = countif(IsClickedThrough == true), FirstClick = min(Timestamp), LastClick = max(Timestamp), ClickedUrls = make_set(Url, 10) by NetworkMessageId; let Remediation = EmailPostDeliveryEvents | where Timestamp > ago(TimeRange) | summarize RemediationActions = make_set(ActionType, 10), LastRemediation = max(Timestamp) by NetworkMessageId, RecipientEmailAddress; Campaigns | join kind=leftouter Delivery on NetworkMessageId, RecipientEmailAddress | join kind=leftouter Clicks on NetworkMessageId | join kind=leftouter Remediation on NetworkMessageId, RecipientEmailAddress | summarize AffectedUsers = dcount(RecipientEmailAddress), Messages = dcount(NetworkMessageId), DeliveredMessages = sum(DeliveredMessages), JunkedMessages = sum(JunkedMessages), BlockedMessages = sum(BlockedMessages), TotalClickEvents = sum(ClickEvents), ClickThroughEvents = sum(ClickThroughEvents), Subjects = make_set(Subjects, 10), SenderDomains = make_set(SenderDomains, 10), ClickedUrls = make_set(ClickedUrls, 10), RemediationActions = make_set(RemediationActions, 10), LastClick = max(LastClick), LastRemediation = max(LastRemediation) by CampaignId, CampaignName, CampaignType, CampaignSubtype | extend SuggestedPriority = case( ClickThroughEvents > 0, "High", TotalClickEvents > 0, "Medium", DeliveredMessages > 0, "Medium", "Low" ) | order by SuggestedPriority asc, AffectedUsers desc, Messages desc This type of query can be useful during hunting sessions, incident review, and campaign reporting. The goal is not only to collect more data. The goal is to help the analyst decide what needs attention first. Correlating Campaign Activity with Microsoft Sentinel When Microsoft Defender XDR is connected to Microsoft Sentinel, incidents and alerts can be synchronized into the Sentinel incident queue. This allows the SOC to correlate campaign-related email activity with other security signals, such as: Suspicious sign-ins Identity alerts Endpoint alerts Cloud app activity OAuth consent activity Data exfiltration attempts Related Microsoft XDR incidents For example, if a user clicked a phishing URL, the SOC can then review whether the same user had suspicious sign-in activity shortly after the click. The following query is a simple starting point for reviewing Microsoft XDR incidents in Microsoft Sentinel. SecurityIncident | where TimeGenerated > ago(30d) | where ProviderName == "Microsoft XDR" | where Title has_any ("phish", "phishing", "email", "malware", "campaign") | summarize Incidents = count(), HighSeverity = countif(Severity == "High"), MediumSeverity = countif(Severity == "Medium"), Closed = countif(Status == "Closed"), Active = countif(Status == "Active") by bin(TimeGenerated, 1d) | order by TimeGenerated desc This query does not replace campaign hunting. It simply helps analysts understand how email-related activity is represented in the Sentinel incident queue. Suggested SOC Workflow A practical campaign-centric workflow could look like this: Step 1: Start from Campaign Views Review campaigns with delivered messages, clicked users, visited links, or high user impact. Step 2: Pivot to KQL Use CampaignInfo to list campaign-related messages and affected recipients. Step 3: Validate Delivery Join with EmailEvents to confirm whether messages were blocked, junked, delivered, or replaced. Step 4: Review User Interaction Join with UrlClickEvents to identify users who clicked URLs or clicked through Safe Links warnings. Step 5: Confirm Remediation Join with EmailPostDeliveryEvents to confirm whether delivered messages were removed after delivery. Step 6: Correlate in Sentinel Review related Microsoft XDR incidents and correlate with identity, endpoint, and cloud activity. Step 7: Decide Response Depending on the impact, the SOC may decide to: Escalate the incident Notify affected users Review user sign-ins Revoke user sessions Reset passwords Block sender domains or URLs Submit false negatives Create a watchlist for related indicators Tune analytics rules or response processes Suggested Priority Logic Not every campaign needs the same level of response. A simple triage model could be: Condition Suggested priority Campaign blocked before delivery Low Campaign delivered to junk Low to Medium Campaign delivered to inbox Medium Campaign delivered to multiple inboxes Medium to High User clicked URL High User clicked through warning High Priority account clicked High Click followed by suspicious sign-in Critical This model should be adapted to each organization’s risk profile and response process. Limitations and Things to Validate Before using this approach in production, validate the following: Defender for Office 365 Plan 2 availability Campaign Views permissions CampaignInfo table availability Defender XDR connector configuration Advanced hunting event streaming Field names in your environment Retention period Data latency Join behavior using NetworkMessageId Whether click events can be joined to email metadata in all cases One important limitation is that some URL click events may not join cleanly with email metadata. For example, clicks from Drafts or Sent Items may not have the same message metadata available for correlation. Also, because CampaignInfo is currently documented as Preview, I would avoid depending on it alone for critical production automation without testing and validation.Pending Approval/Provisioning for Microsoft Defender XDR Lab/Trial Environment
Hello Microsoft Community Team, On June 26, 2026, our organization applied for a Microsoft 365 Developer Environment / Free Trial to support evaluation of the Microsoft Defender XDR Lab environment. To date, the environment has not been provisioned, and we have not received any status updates or confirmation. Impact: Current Status: We are currently utilizing our production environment to test project capabilities, which poses risks and limitations. Future Intent: Our organization plans to transition to a full, paid Business/Enterprise purchase immediately upon proving the platform’s benefits. Urgency: This delay is stalling our evaluation phase. We urgently need this environment onboarded and activated so we can proceed with deployment tests and subsequent procurement. Request: Please review the status of our registration and expedite the onboarding/provisioning of this developer environment. Thank you for your prompt assistance.Sentinel Foundry - MCP Server (Preview) (Github Community Release)
I’ve been cooking something that a lot of people in SOC have been struggling with — especially on the engineering side of Microsoft Sentinel. Thanks to the Microsoft Security team for shaping the capabilities of Sentinel even better with Sentinel Data Lake & Modern SecOps. Today’s the day I can finally share it. Note: This is not an official Microsoft product, but it is designed to make the Sentinel Build even better (complement) with much more intelligence. 🚀 Sentinel Foundry is now in public preview with 43 tools. (Sentinel Foundry - MCP Server) It’s an MCP server built to act like the brain of a strong Sentinel engineer — helping make building, improving, and operating Sentinel far more practical, faster, and honestly more enjoyable. For a lot of teams, the challenge is not understanding what Sentinel can do. The hard part is the engineering work around it: -> Deciding what data should actually be ingested -> Building a clean, scalable Sentinel foundation -> Writing useful detections instead of noisy ones -> Balancing security value with cost -> Turning ideas into deployable engineering outputs That is exactly why I built Sentinel Foundry to help communities grow stronger. It helps with the real engineering tasks behind Sentinel — from architecture thinking to detection design, deployment planning, ingestion strategy, automation ideas, and many of the workflows outlined in the GitHub project. How does it work? Here’s one of the flagship prompts I ran with it: “Give me a complete security posture report for our workspace. Score each pillar and tell me what to prioritise.” And within seconds, it produced a structured engineering blueprint that would normally take a lot longer to pull together manually. You can see the example prompts here in what it can do: https://github.com/prabhukiranveesam/Sentinel-Foundry#what-can-it-do I want building Sentinel to feel less like repetitive engineering overhead — and more like real security engineering that is fast, creative, and enjoyable. If you work with Sentinel as a SOC L2 analyst, engineer, detection engineer, consultant, or architect, I’d genuinely love for you to try it and tell me what you think. 🔗 Public Preview: https://github.com/prabhukiranveesam/Sentinel-Foundry This is just the start of an AI era — and I’m excited to keep shaping it with more powerful features over the coming days. This is very easy to set up and will be available to all of you at no cost during this month as part of the public preview, and your feedback is extremely valuable to shape this as a powerful solution.Exempt a specific container in MDC
I have this recommendation showing in defender. Immutable (read-only) root filesystem should be enforced for containers There are multiple containers inside AKS that are showing as "Unhealthy" airflow/db1 airflow/sql1 airflow/scheduler1 Is there a way to exempt a specific container or the whole recommendation has to be exempted. Thanks
Detecting AI agents and non-human identities in Microsoft Sentinel: the classic-agent blind spot
Build 2026 made the direction official. The industry is moving from the app era into the agent era, and Microsoft spent a real share of the keynote on securing agents across their lifecycle, from discovering what is exploitable to governing what is running in production. On the identity side the centerpiece is Microsoft Entra Agent ID, now generally available, which gives AI agents first-class identities and extends Conditional Access, Identity Protection, and full audit logging to them. That is good news for agents you build the new way. It is not the whole picture, and the gap is where most SOCs will get hurt first. Modern agents are covered. Classic agents are not. Entra Agent ID draws a hard line between two kinds of agent. Modern agents are created through the Agent ID platform, each backed by an agent identity blueprint. They carry a proper Agent ID, a full audit trail, and the complete set of governance capabilities, including Identity Protection for Agents, which establishes a baseline for an agent's normal activity and flags anomalies automatically. Classic agents are everything that came before, or that gets built outside the platform: AI agents implemented as ordinary service principals or app registrations, for example Copilot Studio agents created before Agent ID was enabled, or any home-grown automation calling Graph with client credentials. In the Entra agent registry they appear with "Has Agent ID: No," and that flag matters, because the Agent ID protections apply to identities that actually hold an Agent ID. Classic agents sit outside Identity Protection for Agents and Conditional Access for Agents. Here is the uncomfortable part. The non-human identities you already run, the service principals behind your pipelines, your integrations, your scripts, your pre-platform Copilot Studio bots, are almost all classic agents. They tend to outnumber your human accounts, they have no MFA in any meaningful sense, and a credential added to one does not show up in the Azure portal. The new platform protections do not reach them. Until you migrate them, the only place you get detection coverage on that population is your SIEM. So this is the job Sentinel does that Agent ID does not: detect risky behavior on the classic, service-principal-backed agents that the platform cannot yet protect. The telemetry you have, and the one switch people forget Three tables carry most of the signal. AADServicePrincipalSignInLogs records service principal authentications, the client-credentials sign-ins your agents and automation use. No user, no MFA, just an app proving it holds a secret or certificate. AADManagedIdentitySignInLogs does the same for managed identities. AuditLogs records directory changes, including the one that matters most for persistence: a new credential added to an application or service principal. One practical warning before any of this works. Service principal and managed identity sign-in logs are not streamed by default. You have to enable those categories explicitly in the Entra diagnostic settings feeding your workspace. Plenty of teams write the detection, never check, and never notice the table is empty. Verify that first. Detection 1: a new credential on a service principal or app Adding a secret or certificate to an existing service principal is one of the cleanest persistence techniques in a Microsoft cloud. The attacker compromises a privileged user or app, drops a fresh credential on a service principal that already holds useful Graph permissions, and now has access that survives password resets and session revocation. It maps to MITRE T1098.001, Account Manipulation: Additional Cloud Credentials. For a classic agent it is especially nasty, because there is no Identity Protection baseline watching it. // Detection 1: new secret or certificate added to an application or service principal // MITRE T1098.001 - Account Manipulation: Additional Cloud Credentials AuditLogs | where OperationName has_any ("Add service principal", "Certificates and secrets management") | where Result =~ "success" | extend Initiator = coalesce( tostring(InitiatedBy.user.userPrincipalName), tostring(InitiatedBy.app.displayName)) | extend InitiatorIp = tostring(InitiatedBy.user.ipAddress) | mv-apply Target = TargetResources on ( where Target.type =~ "Application" | extend TargetName = tostring(Target.displayName), TargetId = tostring(Target.id), KeyChanges = Target.modifiedProperties ) | mv-apply Prop = KeyChanges on ( where tostring(Prop.displayName) =~ "KeyDescription" | extend NewKeys = parse_json(tostring(Prop.newValue)), OldKeys = parse_json(tostring(Prop.oldValue)) ) | extend AddedKeys = set_difference(NewKeys, OldKeys) | where array_length(AddedKeys) > 0 | project TimeGenerated, Initiator, InitiatorIp, TargetName, TargetId, AddedKeys | order by TimeGenerated desc The operation filter catches the three shapes this event takes in the log: "Add service principal," "Add service principal credentials," and "Update application - Certificates and secrets management." The modifiedProperties parsing isolates the KeyDescription change, and set_difference confirms a key was actually added rather than removed, so rotating out an old credential does not, on its own, fire the rule. False positives come from legitimate rotation and from automation that provisions app credentials (CI/CD, infrastructure as code). The initiator is the discriminant. A credential added by your deployment pipeline's service account at the usual time is routine. The same change initiated by an interactive admin out of hours, or by an account that never normally touches app credentials, is what you want to surface. Allow-list the expected initiators, not the targets. Detection 2: a classic agent signing in from a first-seen IP A service principal that has only ever authenticated from your Azure regions and suddenly signs in from somewhere new is a strong signal that its credential has been lifted and is being used elsewhere. Service principals have stable, boring network behavior, which makes a first-seen IP a far cleaner indicator for them than it is for roaming human users. This is the behavioral baseline Identity Protection gives you for free on modern agents, rebuilt in KQL for the classic ones it ignores. MITRE T1078.004, Valid Accounts: Cloud Accounts. // Detection 2: classic-agent service principal signing in from a previously unseen IP // MITRE T1078.004 - Valid Accounts: Cloud Accounts let baseline = 14d; let detection = 1d; let KnownIPs = AADServicePrincipalSignInLogs | where TimeGenerated between (ago(baseline + detection) .. ago(detection)) | where tostring(ResultType) == "0" | summarize KnownIPSet = make_set(IPAddress) by AppId; AADServicePrincipalSignInLogs | where TimeGenerated > ago(detection) | where tostring(ResultType) == "0" | lookup kind=leftouter KnownIPs on AppId | where set_has_element(KnownIPSet, IPAddress) == false | summarize FirstSeen = min(TimeGenerated), Resources = make_set(ResourceDisplayName, 10) by ServicePrincipalName, AppId, IPAddress | order by FirstSeen desc The query builds a per-application baseline of source IPs over the previous two weeks, then flags any successful sign-in today from an address outside that set. Two tuning notes. Brand-new service principals have no baseline, so they surface on first use. That is usually worth seeing once, but you can exclude AppIds younger than the baseline window if it gets noisy. And if your agents egress through shifting cloud IP ranges, widen the comparison from an exact IP to the autonomous system number or a known-range allow-list, otherwise you will chase your own infrastructure. This complements Agent ID, it does not replace it! The endgame is not to run these rules forever. It is to shrink the population they apply to. Inventory your tenant for agents marked "Has Agent ID: No," prioritize the ones holding sensitive Graph permissions, and migrate them onto the Agent ID platform, where Identity Protection and Conditional Access take over the baselining you are doing here by hand. Microsoft has signaled a migration path from classic to modern agents. Treat these two detections as the coverage you need in the meantime, and as a permanent safety net for anything that never makes the move. If you do one thing this week: enable the service principal sign-in log category, deploy detection 1, and pull a list of every service principal that had a credential added in the last 90 days. That list alone tends to be more interesting than people expect. Cheers, Marcel
Performance in scanning
We are trying to search for CUI data on internal file stores. Last week, I decided to run another discovery scan, this time using ALL instead of Policy Only. It took much longer and left the scanner server in an almost unusable state and didn’t give really any more information than the first one did. Based on my research, we need to define and set the policy before we run scans. This is the information tip from the Purview scanner settings: Scan started at: 2026-05-20 22:54:06Z Scan ended at: 2026-05-24 16:16:51Z Scan duration: 3 days, 17 hours, 22 minutes, 45 seconds Scan id: 93acb922-e2ac-4fb7-b259-d6184e7aa434 Repository: \\cab-filesrv-01.fg.com\Departments. Enforce mode is Off Scanned files:3509640 Actions: Classified:3369456 Classified as Public:14 Classified as Fg Private:3369442 Labeled:0 Remove label:0 Protected:0 Remove protection:0 Files with matched information types:572895 Skipped due to - No match:0 Skipped due to - Not supported:0 Skipped due to - Already labeled:0 Skipped due to - Already scanned:0 Skipped due to - Require justification:0 Skipped due to - Unknown reason:0 Skipped due to - Excluded:98833 Skipped due to - Attribute:0 Failed:41318Purview DLP Behaviours in Outlook Desktop
We are currently testing Microsoft Purview DLP policies for user awareness, where sensitive information shared externally triggers a policy tip, with override allowed (justification options enabled) and no blocking action configured. We are observing the following behaviours in Outlook Desktop: Inconsistent policy tip display (across Outlook Desktop Windows clients) – For some users, the policy tip renders correctly, while for others it appears with duplicated/stacked lines of text. This is occurring across users with similar configurations. Override without justification – Users are able to click “Send Anyway/Confirm and send” without selecting any justification option (e.g. business justification, manager approval, etc.), which bypasses the intended control. New Outlook: Classic Outlook: This has been observed on Outlook Desktop (Microsoft 365 Apps), including: Version 2602 (Build 19725.20170 Click-to-Run) Version 2602 (Build 16.0.19725.20126 MSO) Has anyone experienced similar behaviour with DLP policy tips or override enforcement in Outlook Desktop? Keen to understand if this is a known issue or if there are any recommended fixes or workarounds.
I just want to secure AI. DLP vs Info Protection vs DSPM vs Governance vs...
I'm with an MSP, and I've avoided Purview like the plague, because it seems to be suffering from the same 'made by marketing teams' 'strategy' the 365 documentation is. However, it's my understanding Purview policies are needed for Data control of Copilot. Here's my issue: all of these different 'solutions' sound like the exact same thing, but are pitched as if they are something different. i'm going to post a couple of descriptions for these 'solutions' to illustrate this. 'discover, label, and protect sensitive and business-critical info' 'make sure your organization can identify, monitor, and protect sensitive info across the expanding Microsoft 365 landscape' 'discover and secure all your sensitive data across Microsoft 365 and non-365 data sources' 'Discover, label, and protect sensitive and business-critical info across your multicloud data estate.' I genuinely do not have time to figure out what each of these 'solutions' are, then figure out their policies, then their giant library of settings (below)... It's not even clear to me what's active NOW, considering we never licensed Purview - but somehow have been roped into it. It SEEMS like these are all variations of marketing terms, which all point to 3-4 actual technical implementations in obscure ways. Can someone advise on the ACTUAL technical policies we want to target and enable? Or just give some clarity? I've never felt so overwhelmed or disconnected from Microsoft's environment. We just want to secure our tenant's AI usage.Unified Catalog Self-serve analytics integration
I'm hoping someone has gone through the process of setting up the Self-serve analytics in the Unified Catalog settings to push the Unified Catalog information down to a Fabric Lakehouse. I created a Workspace, and then created a lakehouse in this workspace, and created a folder under the files section in the lakehouse. I used the MSI that is shown in Purview when you configure the storage for the connection and granted it contriubutor access to the Workspace. I then went into Purview, settings for Unified Catalog, and in the solution integrations, set up Fabric storage and provided the URL to the File folder I set up on the lakehouse. I tested the connection and it tested successfully. When I set up the scheduler to run, I received the following: The blacked out is the Workspace ID. I'm trying to understand what I'm missing, I'm assuming write permissions are missing somewhere, but I'm not sure. Any assistance is appreciated.
Events
in 6 hours
Modern identity security means stopping attacks before they escalate and extending protection beyond human users to apps and agentic identities across your identity fabric.
Learn how Microsoft Entr...
Online
Join us for a live demo and AMA on the Microsoft Sentinel SIEM migration experience. We’ll show how the experience helps teams move from legacy SIEMs like Splunk and QRadar into Microsoft Sentinel wi...
Online
