Detections
Explore public detection logic contributed by the community across SIEM and rule languages.
3,142 detections
Filters
Last updated
All Time
Detection languages
2,293
246
196
179
104
Contributors
738
670
569
284
148
Categories
1,164
747
693
623
440
Platforms
39,960
7,270
6,537
3,868
3,651
3,142
Products / Services
2,850
742
269
187
182
MITRE Techniques
872
857
778
768
738
CVEs
13
5
2
1
1
IDS Classtypes
3
2
2
2
1
IDS Protocols
9
1
1
Detects when a single user deletes more than 25 emails within a 2-minute window. This behavior may indicate a compromised account, insider threat activity, data destruction attempt, or automated malicious script. The detection monitors Exchange operations including MoveToDeletedItems, SoftDelete, and HardDelete actions.
Detects possible Adversary-in-the-Middle (AiTM) attacks, session cookie replay, MFA session hijacking, token theft, and stolen cloud session reuse by identifying abnormal SessionId reuse across multiple IPs, browsers, and devices. This detection compares current session activity against a 14-day historical user behavior baseline to identify new or abnormal behavior. It includes risk scoring and account classification, while reducing false positives from token refresh activity and common Microsoft Teams noise.
This rule tracks user and administrator submissions of potential phishing emails to Microsoft for analysis within the Security & Compliance Center. It extracts and logs details such as the submission verdict, rescan verdict, recipients, subject, content, submitter's display name, sender information (P1 and P2), email destination, final delivery verdict, and sender IP address. This provides visibility into an organization's proactive efforts to report and analyze suspected phishing attempts.
High Email Volume to Personal Accounts
Cortex XDR
Detects potential data exfiltration or policy violation by identifying users sending a high volume of emails, large emails, or emails to many unique personal email addresses (e.g., Gmail, Yahoo, Outlook.com) from an Office 365 Exchange Online mailbox. This rule aggregates email activity by sender and flags if any of the defined thresholds are met: total sent data to personal addresses exceeds 25 MB, total messages sent to personal addresses exceeds 10, emails sent to more than 5 unique personal addresses, or any single message sent to a personal address exceeds 25 MB. This is designed to be ran as a scheduled correlation for a day time frame. Thresholds can also be configured to fit your environment. Each user will have a different line that contains all of the matching emails they sent to, the attachments, and subjects of the emails.
SharePoint Online Sensitive Data Search
Cortex XDR
This rule detects when users perform search queries in SharePoint Online that contain keywords indicative of sensitive data. The rule categorizes these searches into several types: credentials, personnel/HR information, infrastructure reconnaissance, financial/PII, and confidential business information.
Creating multiple detections without reviewing them over time to ensure they’re still effective can easily turn them into pure noise, adding nothing but fatigue to our day-to-day incident analysis.
The DOVE Model (Detection Overlap & Value Evaluation) evaluates detections across four main dimensions (https://zenodo.org/records/20011459):
1. Detection Type
IOC-based detections tend to have a shorter lifecycle, current TI sources are collecting a huge number of indicators based on new attack or threats which present a higher duplication risk.
IOA-based detections focus on behaviors, have a longer lifecycle, provide higher unique value and requires extended time to create corresponding detection.
2. System / Technology Coverage
Mainstream technologies (e.g., Windows, M365, macOS) are heavily monitored and widely covered by SIEM platforms, increasing duplication probability.
Non-mainstream or custom systems offer less coverage and therefore provide opportunities for more unique detections.
3. Threat Recency
Older or well-known threats are widely documented and likely already covered by existing detections.
New or emerging threats are less likely to be covered, reducing duplication risk and increasing detection value.
4. Source / Provider
Built-in or native detections are already integrated into SIEM platforms and often overlap with custom logic.
Custom or third-party detections introduce tailored logic and reduce overlap probability.
The following KQL query can evaluate how useful are our current detections with the aim to detect alerts that are showing the same assets within the same hour, which indicates potential overlap.
Custom detections added in current SIEM or XDR solutions can easily turn from useful detections into noise, so analyzing their results is crucial.
The DOVE Model (Detection Overlap & Value Evaluation) evaluates detections across four main dimensions (https://zenodo.org/records/20011459):
1. Detection Type
IOC-based detections tend to have a shorter lifecycle, current TI sources are collecting a huge number of indicators based on new attack or threats which present a higher duplication risk.
IOA-based detections focus on behaviors, have a longer lifecycle, provide higher unique value and requires extended time to create corresponding detection.
2. System / Technology Coverage
Mainstream technologies (e.g., Windows, M365, macOS) are heavily monitored and widely covered by SIEM platforms, increasing duplication probability.
Non-mainstream or custom systems offer less coverage and therefore provide opportunities for more unique detections.
3. Threat Recency
Older or well-known threats are widely documented and likely already covered by existing detections.
New or emerging threats are less likely to be covered, reducing duplication risk and increasing detection value.
4. Source / Provider
Built-in or native detections are already integrated into SIEM platforms and often overlap with custom logic.
Custom or third-party detections introduce tailored logic and reduce overlap probability.
The following KQL query can evaluate how useful are our current detections with the aim to detect alerts that are showing the same assets within the same hour, which indicates potential overlap.
Custom detections added in current SIEM or XDR solutions can easily turn from useful detections into noise, so analyzing their results is crucial.
Detects users accessing, downloading, or copying a high volume of document files (>100 unique files) across multiple SharePoint sites (>1) within a single hour, excluding legitimate OneDrive sync client activity. The rule targets file types commonly associated with sensitive business data — Office documents, PDFs, archives, databases, email exports (PST/OST/MSG), and OneNote notebooks — and aggregates by user and hour to surface bulk collection patterns consistent with insider data theft, account compromise, or pre-exfiltration staging by ransomware affiliates
MC1293480: https://admin.cloud.microsoft/?ref=MessageCenter/:/messages/MC1293480
This rule detects the creation or modification of Exchange inbox rules with highly suspicious or obfuscated names. Adversaries often create inbox rules with short, non-descriptive, or special character names (e.g., '.', '...', ';') to forward, redirect, or delete emails, facilitating data exfiltration or maintaining persistence. The rule specifically looks for rule names that are very short (less than 3 characters), empty, or match a predefined list of suspicious names.
Identifies emails spoofing NIPR, NAIC, or Stripe domains with invoice-related lures that fail authentication. This rule specifically looks for emails where the sender's domain (SenderFromDomain or SenderMailFromDomain) is one of the targeted domains (nipr.com, naic.org, stripe.com), the subject contains keywords like 'invoice' or 'past due', and the email authentication details do not indicate a 'pass'. This combination suggests a phishing attempt leveraging brand impersonation and financial urgency.
This rule detects successful sign-ins to specific Microsoft 365 applications (identified by ApplicationId '08ad6f98-a4f8-4635-bb8d-f1a3044760f0' or '07c030f6-5743-41b7-ba00-0a6e85f37c17') from devices that have previously established a network connection with a process named 'cowork-svc'. The rule correlates network connection events with Azure AD sign-in events over a 30-day lookback period to identify potentially legitimate or suspicious activity related to 'cowork-svc' application usage.
Detects signed executable files with untrusted certificates (expired, revoked,
or broken chain) that were downloaded from the internet and executed within one
hour of being written to a high-risk path such as user temp, AppData, Downloads,
or Public directories.
An untrusted certificate on a downloaded executable can indicate use of a
revoked or stolen code-signing certificate, a technique observed in multiple
Iranian APT campaigns including Seedworm/MuddyWater (ref: CyberProof, April 2026)
and consistent with CISA advisory AA26-097A.
Note: IsTrusted reflects certificate status at time of MDE cloud scan, not
real-time revocation status. Files executed before MDE scan completion or before
CA revocation may not be detected.
or broken chain) that were downloaded from the internet and executed within one
hour of being written to a high-risk path such as user temp, AppData, Downloads,
or Public directories.
An untrusted certificate on a downloaded executable can indicate use of a
revoked or stolen code-signing certificate, a technique observed in multiple
Iranian APT campaigns including Seedworm/MuddyWater (ref: CyberProof, April 2026)
and consistent with CISA advisory AA26-097A.
Note: IsTrusted reflects certificate status at time of MDE cloud scan, not
real-time revocation status. Files executed before MDE scan completion or before
CA revocation may not be detected.
If you are in a country where there is more than one local language, like me in Switzerland, it is challenging to create detections based on subjects because you can have them in multiple languages. Therefore, focusing on the value odspnotify inside of InternetMessageId - which does not change based on the language - helps to detect cases.
This one is related to OneTimePassCode and is useful to identify if the code is sent to other domains, which could mean that a forwarding rule has been configured or some other suspicious activity.
This one is related to OneTimePassCode and is useful to identify if the code is sent to other domains, which could mean that a forwarding rule has been configured or some other suspicious activity.
The are some cases where InternetMessageIds contain a date integrated, sometimes encoded, sometimes not. Therefore, we have two timestamps to compare:
- InternetMessageId date → usually when the sending system generated the message.
- Timestamp (EmailEvents) → when Microsoft received/processed the email.
So if we compare “when it was created” vs “when it arrived”, we can detect cases such as emails created by the system but, due to local configuration, email campaigns that deliver slowly, or messages that were postponed compared to when they were received by Microsoft.
The ratio of true positive was really high and multiple emails were not detected as a Threat so good case to add as a detection.
- InternetMessageId date → usually when the sending system generated the message.
- Timestamp (EmailEvents) → when Microsoft received/processed the email.
So if we compare “when it was created” vs “when it arrived”, we can detect cases such as emails created by the system but, due to local configuration, email campaigns that deliver slowly, or messages that were postponed compared to when they were received by Microsoft.
The ratio of true positive was really high and multiple emails were not detected as a Threat so good case to add as a detection.
I don't know how many time I suffered malicious actors sharing links via Microsoft services such as SharePoint or other sites manipulating them for phishing, malware, or other malicious purposes.
I have seen multiple detections to detect phishing campaigns based on subjects, such as detecting when an email subject ends with “with you” or contains “shared.” However, if the original mailbox is configured in another language, these rules become useless.
This change specifying the InternetMessageID value odspnotify and, in addition, specifying the first part of the InternetMessageId, which tells us the action requested or performed by the email—in this case, share.
For this detection, I recommend adding your trusted domains, but it is incredibly useful to detect when links from Microsoft systems are shared with your users from unexpected domains.
I have seen multiple detections to detect phishing campaigns based on subjects, such as detecting when an email subject ends with “with you” or contains “shared.” However, if the original mailbox is configured in another language, these rules become useless.
This change specifying the InternetMessageID value odspnotify and, in addition, specifying the first part of the InternetMessageId, which tells us the action requested or performed by the email—in this case, share.
For this detection, I recommend adding your trusted domains, but it is incredibly useful to detect when links from Microsoft systems are shared with your users from unexpected domains.
https://www.microsoft.com/en-us/security/blog/2026/04/18/crosstenant-helpdesk-impersonation-data-exfiltration-human-operated-intrusion-playbook/
Detects LOLBins (cmd.exe, powershell.exe, rundll32.exe, mshta.exe, etc.) spawned
directly by a web browser, indicative of FileFix or ClickFix-style social engineering
execution.
In these attacks, a malicious webpage silently copies a disguised command
to the user's clipboard; the user is then instructed to paste it into the Windows File
Explorer address bar or Run dialog, resulting in LOLBin execution under the browser's
process context.
Observed delivering loaders, RATs, and ransomware precursors in the
wild since July 2025.
Exclusions applied for Chrome Native Messaging IPC, ClickOnce
application launchers, and OAuth2/OIDC SSO token redirect fragments.
directly by a web browser, indicative of FileFix or ClickFix-style social engineering
execution.
In these attacks, a malicious webpage silently copies a disguised command
to the user's clipboard; the user is then instructed to paste it into the Windows File
Explorer address bar or Run dialog, resulting in LOLBin execution under the browser's
process context.
Observed delivering loaders, RATs, and ransomware precursors in the
wild since July 2025.
Exclusions applied for Chrome Native Messaging IPC, ClickOnce
application launchers, and OAuth2/OIDC SSO token redirect fragments.
Detects inbound network connections and Azure AD sign-in attempts originating from the 21-IP RDP scanning fleet attributed to AS213438 (ColocaTel Inc., Seychelles), as reported by GreyNoise on April 10, 2026. The fleet generated 67.4% of global RDP Crawler activity on April 7, 2026 across four /24 blocks geolocated to the Netherlands. The same ASN demonstrated a burst-and-crash pattern in both March and April 2026, suggesting further rotations are likely.
Detects the creation or modification of inbox rules that move messages to rarely checked folders like Archive, RSS Subscriptions, Conversation History, Junk Email, or Deleted Items. This is a common post-exploitation technique used to hide fraudulent communications (BEC, payroll fraud) or suppress security alerts, MFA notifications, and password reset emails from the victim. Approximately 10% of compromised accounts have malicious mailbox rules created shortly after initial access.
Detects when a Microsoft Cloud App Security reported a risky sign-in attempt due to a login associated with an impossible travel.
Detects instances where an email, identified as malicious or suspicious by the Microsoft Defender for Office 365 (formerly ATP) engine, was delivered to a user's Inbox or Junk folder.
It might indicate that a potential threat, such as a spearphishing attachment or links, has bypassed initial blocking mechanisms and reached an end-user, requiring further investigation and potential remediation.
It might indicate that a potential threat, such as a spearphishing attachment or links, has bypassed initial blocking mechanisms and reached an end-user, requiring further investigation and potential remediation.
Page 113 of 150