Detections
Explore public detection logic contributed by the community across SIEM and rule languages.
40,407 detections
Filters
Last updated
All Time
Detection languages
15,618
13,637
2,824
1,803
1,727
Contributors
7,678
5,999
5,315
4,504
4,206
Categories
18,287
9,897
3,918
3,796
3,721
Platforms
40,407
7,473
6,580
3,935
3,713
Products / Services
10,635
9,722
6,762
1,902
1,895
MITRE Techniques
13,801
13,234
8,133
6,014
4,458
CVEs
49
45
30
30
29
IDS Classtypes
214
56
43
23
19
IDS Protocols
183
172
20
16
8
Detects the correlation of pass-the-hash execution tooling (such as Invoke-TheHash, Invoke-SMBExec, or Invoke-WMIExec) followed by the execution of the GodPotato privilege escalation exploit on the same device within a 24-hour timeframe.
Detects the execution of a fake certificate-fix MSI package via msiexec.exe or the creation of CertFixer.msi, followed within one hour by the creation of a Run key persistence mechanism named 'WindowsUpdateAssist' on the same device.
Detects HTTP requests made to specific fabricated decoy routes planted as comments in application code or documentation (SpikeStrip comments). Probing these non-existent routes indicates that an agent or adversary has ingested the planted comments and is actively scanning the web application.
Detects execution of tunneling, proxy, and remote exposure tools associated with unauthorized access and protocol tunneling and web sockets.
Detects the presence of "UWhRC....AAYBAAAA" pattern in command line.
The pattern "1UWhRCAAAAA..BAAAA" is a base64-encoded signature that corresponds to a marshaled CREDENTIAL_TARGET_INFORMATION structure.
Attackers can use this technique to coerce authentication from victim systems to attacker-controlled hosts.
It is one of the strong indicators of a Kerberos coercion attack, where adversaries manipulate DNS records
to spoof Service Principal Names (SPNs) and redirect authentication requests like in CVE-2025-33073.
If you see this pattern in the command line, it is likely an attempt to add spoofed Service Principal Names (SPNs) to DNS records,
or checking for the presence of such records through the `nslookup` command.
The pattern "1UWhRCAAAAA..BAAAA" is a base64-encoded signature that corresponds to a marshaled CREDENTIAL_TARGET_INFORMATION structure.
Attackers can use this technique to coerce authentication from victim systems to attacker-controlled hosts.
It is one of the strong indicators of a Kerberos coercion attack, where adversaries manipulate DNS records
to spoof Service Principal Names (SPNs) and redirect authentication requests like in CVE-2025-33073.
If you see this pattern in the command line, it is likely an attempt to add spoofed Service Principal Names (SPNs) to DNS records,
or checking for the presence of such records through the `nslookup` command.
Detects execution of Advanced Installer MSIX Package Support Framework (PSF) components, specifically AI_STUBS executables with original filename 'popupwrapper.exe'.
This activity may indicate malicious MSIX packages build with Advanced Installer leveraging the Package Support Framework to bypass application control restrictions.
This activity may indicate malicious MSIX packages build with Advanced Installer leveraging the Package Support Framework to bypass application control restrictions.
Detects rare, non-system, non-browser processes opening a TCP listening connection on port 56793, which is associated with the rat_http3_cc embedded local HTTP server. Excludes processes that have a baseline of being seen listening on this port across multiple devices.
Detects Node-RED admin API flow deployments (POST /flows) where the payload contains an 'http in' node, a 'function' node, an 'exec' node, and 'addpay' enabled, which can indicate malicious workflow deployment or command execution attempts.
Self deletion through timeout and del
Cortex XDR
Detects command prompt (cmd.exe) executing a command containing 'timeout', 'del', and '/t', commonly used by scripts or malware to self-delete after execution by pausing via timeout and then deleting the file.
Detects when a process writes a debug log file named volta_debug.txt into the AppData directory and subsequently deletes itself, a behavior associated with the Volta infostealer attempting to clean up traces.
Detects creation of archive files in temp directories (Sysmon Event ID 11). A single event is the unit of behaviour; the burst relationship (more than 3 archives by the same Image/host within 10 minutes for medium environments, >2 small, >5 large) must be applied via a correlation/threshold in the SIEM, e.g. count of TargetFilename grouped by Computer and Image over a 10 minute window, ahead of cloud storage exfiltration.
Detects a single host initiating WinRM connections (TCP 5985/5986) to many
distinct internal hosts within a compressed window. Autonomous AI agents
performing lateral movement fan out to multiple targets simultaneously —
testing credentials and executing commands across the network in parallel
rather than the sequential, one-host-at-a-time pattern of human operators.
ARTEX worker agents receive credential sets from the exploration graph and
immediately attempt WinRM access to every reachable Windows host. The fan-
out pattern — many simultaneous outbound connections on port 5985/5986
from a single source — is unusual for legitimate administration where even
PowerShell remoting scripts process hosts sequentially.
Sysmon Event ID 3 (network connection) with DestinationPort 5985/5986
provides the telemetry. Adjust the destination count threshold based on
typical remote administration patterns.
BQL (Bloo Query Language):
stream=ep-network where action='NETWORK_CONNECTION'
and DstPort in (5985, 5986)
and DstIP rlike '^(10\.|192\.168\.|172\.(1[6-9]|2[0-9]|3[01])\.)'
| groupby SrcIP, DstIP
| having count > 5
distinct internal hosts within a compressed window. Autonomous AI agents
performing lateral movement fan out to multiple targets simultaneously —
testing credentials and executing commands across the network in parallel
rather than the sequential, one-host-at-a-time pattern of human operators.
ARTEX worker agents receive credential sets from the exploration graph and
immediately attempt WinRM access to every reachable Windows host. The fan-
out pattern — many simultaneous outbound connections on port 5985/5986
from a single source — is unusual for legitimate administration where even
PowerShell remoting scripts process hosts sequentially.
Sysmon Event ID 3 (network connection) with DestinationPort 5985/5986
provides the telemetry. Adjust the destination count threshold based on
typical remote administration patterns.
BQL (Bloo Query Language):
stream=ep-network where action='NETWORK_CONNECTION'
and DstPort in (5985, 5986)
and DstIP rlike '^(10\.|192\.168\.|172\.(1[6-9]|2[0-9]|3[01])\.)'
| groupby SrcIP, DstIP
| having count > 5
Detects a single host deploying 2 or more distinct persistence techniques (Run key write, startup folder file drop, scheduled task creation, service creation) within 10 minutes. Sigma has no native sequence/aggregation, so this single rule matches each persistence technique individually; the 2-or-more-distinct-techniques-in-10-minutes logic must be applied in the SIEM (see comments at the end of the rule).
Detects a burst of many distinct DNS queries from one host within a 5-minute window, exceeding human tempo,
consistent with autonomous-agent subdomain enumeration (reconnaissance). Sigma core cannot count distinct values,
so this rule selects the Sysmon DNS query events (Event ID 22) and must be paired with a backend-side correlation:
count distinct QueryName per Computer (or Image) over timeframe 5m, threshold >50 small / >80 medium / >150 large environments.
Enrich against asset inventory to exclude known scanners and build servers (Tier C leading indicator).
consistent with autonomous-agent subdomain enumeration (reconnaissance). Sigma core cannot count distinct values,
so this rule selects the Sysmon DNS query events (Event ID 22) and must be paired with a backend-side correlation:
count distinct QueryName per Computer (or Image) over timeframe 5m, threshold >50 small / >80 medium / >150 large environments.
Enrich against asset inventory to exclude known scanners and build servers (Tier C leading indicator).
Detects process creation of AD/host/network discovery commands (T1087.002, T1018) spawned from a scripting or shell parent (T1059), the building block of an autonomous agent's enumerate-exploit-enumerate loop.
Sigma cannot count events on its own: aggregate this rule per host with a threshold of 15 or more matching events (discovery plus execution) in a 30 minute window, and require both a discovery selection and an execution-parent selection to appear in the window to capture the cyclical pattern. Response per the report: isolate the host and alert the SOC.
Sigma cannot count events on its own: aggregate this rule per host with a threshold of 15 or more matching events (discovery plus execution) in a 30 minute window, and require both a discovery selection and an execution-parent selection to appear in the window to capture the cyclical pattern. Response per the report: isolate the host and alert the SOC.
Detects a single source IP attempting authentication against multiple
distinct service categories (VPN, webmail, VDI, RDP, cloud console, API
gateway) within a compressed time window. Human attackers focus on one
vector at a time — an autonomous AI framework fans credential tests across
every available authentication surface simultaneously.
ARTEX worker agents receive a credential from the exploration graph and
immediately test it against all discovered authentication endpoints in
parallel. The resulting pattern — the same source hitting SSH, OWA, VPN,
and cloud login within seconds — is distinctive and rare in legitimate
traffic.
This rule requires correlation across authentication log sources. Deploy
with a SIEM that normalizes authentication events across network,
endpoint, and cloud sources.
BQL (Bloo Query Language):
stream=authentication where action='LOGIN' and status='FAILED'
| groupby source_ip, destination_ip
| having count > 10
distinct service categories (VPN, webmail, VDI, RDP, cloud console, API
gateway) within a compressed time window. Human attackers focus on one
vector at a time — an autonomous AI framework fans credential tests across
every available authentication surface simultaneously.
ARTEX worker agents receive a credential from the exploration graph and
immediately test it against all discovered authentication endpoints in
parallel. The resulting pattern — the same source hitting SSH, OWA, VPN,
and cloud login within seconds — is distinctive and rare in legitimate
traffic.
This rule requires correlation across authentication log sources. Deploy
with a SIEM that normalizes authentication events across network,
endpoint, and cloud sources.
BQL (Bloo Query Language):
stream=authentication where action='LOGIN' and status='FAILED'
| groupby source_ip, destination_ip
| having count > 10
Detects the creation of multiple compressed archive files (.zip, .7z,
.tar.gz, .rar) in temporary or staging directories within a compressed
window. Autonomous AI agents performing collection stage data for
exfiltration by creating archives at machine speed — producing multiple
compressed files in seconds rather than the occasional single archive a
human creates.
ARTEX worker agents tasked with data collection compress discovered files
into archives in temp directories before exfiltration. The exploration
graph tracks which files have been staged, and the planner dispatches
parallel workers to compress different data types simultaneously. The
resulting burst of archive file creation events in Sysmon Event ID 11 is
distinctive.
Monitor for archive creation in %TEMP%, %APPDATA%, C:\ProgramData, and
/tmp. Correlate with subsequent outbound network connections from the same
host.
BQL (Bloo Query Language):
stream=ep-file where action='FILE_CREATED'
and (TargetFilename like '%.zip' or TargetFilename like '%.7z'
or TargetFilename like '%.tar.gz' or TargetFilename like '%.rar')
and (TargetFilename like '%\\Temp\\%' or TargetFilename like '%\\tmp\\%'
or TargetFilename like '%\\ProgramData\\%')
| groupby system
| having count > 3
.tar.gz, .rar) in temporary or staging directories within a compressed
window. Autonomous AI agents performing collection stage data for
exfiltration by creating archives at machine speed — producing multiple
compressed files in seconds rather than the occasional single archive a
human creates.
ARTEX worker agents tasked with data collection compress discovered files
into archives in temp directories before exfiltration. The exploration
graph tracks which files have been staged, and the planner dispatches
parallel workers to compress different data types simultaneously. The
resulting burst of archive file creation events in Sysmon Event ID 11 is
distinctive.
Monitor for archive creation in %TEMP%, %APPDATA%, C:\ProgramData, and
/tmp. Correlate with subsequent outbound network connections from the same
host.
BQL (Bloo Query Language):
stream=ep-file where action='FILE_CREATED'
and (TargetFilename like '%.zip' or TargetFilename like '%.7z'
or TargetFilename like '%.tar.gz' or TargetFilename like '%.rar')
and (TargetFilename like '%\\Temp\\%' or TargetFilename like '%\\tmp\\%'
or TargetFilename like '%\\ProgramData\\%')
| groupby system
| having count > 3
Detects a single host issuing a high volume of LDAP search queries targeting
user, group, computer, and organizational unit objects within a compressed
window. Autonomous AI agents perform Active Directory enumeration at machine
speed — systematically querying objectClass=user, objectClass=group,
objectClass=computer, and servicePrincipalName attributes in rapid succession
to build a complete domain graph.
This mirrors BloodHound-style collection but at velocities that exceed what
SharpHound or manual LDAP browsing produce. ARTEX feeds LDAP results into
its dual-graph architecture (asset graph + exploration graph), enabling
immediate path-finding from enumeration to privilege escalation without
human review.
Windows Event ID 1644 must be enabled via LDAP diagnostics registry key
(HKLM\SYSTEM\CurrentControlSet\Services\NTDS\Diagnostics, Field Engineering = 5).
BQL (Bloo Query Language):
stream=win-audit where event_id='1644'
and not source_ip rlike '^127\.'
| groupby source_ip
| having count > 50
user, group, computer, and organizational unit objects within a compressed
window. Autonomous AI agents perform Active Directory enumeration at machine
speed — systematically querying objectClass=user, objectClass=group,
objectClass=computer, and servicePrincipalName attributes in rapid succession
to build a complete domain graph.
This mirrors BloodHound-style collection but at velocities that exceed what
SharpHound or manual LDAP browsing produce. ARTEX feeds LDAP results into
its dual-graph architecture (asset graph + exploration graph), enabling
immediate path-finding from enumeration to privilege escalation without
human review.
Windows Event ID 1644 must be enabled via LDAP diagnostics registry key
(HKLM\SYSTEM\CurrentControlSet\Services\NTDS\Diagnostics, Field Engineering = 5).
BQL (Bloo Query Language):
stream=win-audit where event_id='1644'
and not source_ip rlike '^127\.'
| groupby source_ip
| having count > 50
Detects expensive/inefficient LDAP search queries logged in Directory Service Event 1644 whose filters target user, group or computer object discovery (T1087.002, T1069.002). Machine-velocity enumeration should be identified by aggregating matching events per ClientIP over a short window (SIEM correlation, e.g. Sigma value_count/event_count correlation on ClientIP). Requires NTDS Diagnostics '15 Field Engineering' = 5 under HKLM\SYSTEM\CurrentControlSet\Services\NTDS\Diagnostics, otherwise Event 1644 is not generated.
Detects modification of Windows Firewall rules within minutes of a new
remote logon session being established. Autonomous AI agents performing
defense evasion programmatically modify host firewall rules to open
additional ports for C2, lateral movement, or exfiltration immediately
after gaining access — before a human defender can observe the initial
compromise.
ARTEX worker agents tasked with post-exploitation routinely disable
firewall profiles or add allow rules for tools they need (reverse shells,
proxy listeners, tunneling software). The correlation between a fresh
logon (Event 4624 Type 3 or 10) and subsequent firewall modification
(Event 4946/4947 or netsh/Set-NetFirewallProfile execution) within a
compressed window is the detection anchor.
Requires Windows Security Event 4624 and Sysmon Event ID 1 for process
command line capture.
BQL (Bloo Query Language):
stream=ep-process where action='PROCESS_ADDED'
and (lower(image) like '%\\netsh.exe'
or lower(image) like '%\\powershell.exe'
or lower(image) like '%\\pwsh.exe')
and (lower(commandline) like '%firewall%'
or lower(commandline) like '%advfirewall%'
or lower(commandline) like '%set-netfirewallprofile%'
or lower(commandline) like '%new-netfirewallrule%')
remote logon session being established. Autonomous AI agents performing
defense evasion programmatically modify host firewall rules to open
additional ports for C2, lateral movement, or exfiltration immediately
after gaining access — before a human defender can observe the initial
compromise.
ARTEX worker agents tasked with post-exploitation routinely disable
firewall profiles or add allow rules for tools they need (reverse shells,
proxy listeners, tunneling software). The correlation between a fresh
logon (Event 4624 Type 3 or 10) and subsequent firewall modification
(Event 4946/4947 or netsh/Set-NetFirewallProfile execution) within a
compressed window is the detection anchor.
Requires Windows Security Event 4624 and Sysmon Event ID 1 for process
command line capture.
BQL (Bloo Query Language):
stream=ep-process where action='PROCESS_ADDED'
and (lower(image) like '%\\netsh.exe'
or lower(image) like '%\\powershell.exe'
or lower(image) like '%\\pwsh.exe')
and (lower(commandline) like '%firewall%'
or lower(commandline) like '%advfirewall%'
or lower(commandline) like '%set-netfirewallprofile%'
or lower(commandline) like '%new-netfirewallrule%')
Detects process creation where native search tools are used to hunt for plaintext credentials in files (T1552.001) or for private key files (T1552.004). The source report supplies no specific commands, so this is a single-event rule. Burst or velocity correlation per host has to be added in the SIEM.
Page 15 of 1925