Detections
Explore public detection logic contributed by the community across SIEM and rule languages.
3,647 detections
Filters
Last updated
All Time
Detection languages
1,768
374
351
273
255
Contributors
908
676
593
359
164
Categories
1,502
1,169
782
528
521
Platforms
39,936
7,266
6,537
3,866
3,647
Products / Services
807
512
432
340
300
MITRE Techniques
1,180
582
528
506
438
CVEs
17
16
9
8
7
IDS Classtypes
41
15
10
6
5
IDS Protocols
47
24
3
2
1
Detects the execution of processes named 'nohup' with arguments suggesting renderer masquerading ('--type=renderer') containing suspicious terms like 'SystemUpdate' or 'iSync', originating from Cursor IDE helper processes or the terminal shell. This behavior is indicative of a backdoor or malicious process attempting to hide in plain sight by masquerading as legitimate IDE rendering components.
Detects the registration or installation of an MCP (Model Context Protocol) server or agent tool containing suspicious prompt-injection strings (e.g., instructions to ignore safety guardrails or role overrides), immediately followed by the execution of shell commands or unauthorized network connections originating from that tool.
Detects unauthorized modifications to AI agent configuration files, including system prompts, tool allow-lists, and MCP server registration files. Such changes are often performed by processes outside of approved configuration management or infrastructure-as-code pipelines, potentially allowing for persistent manipulation of AI model behavior and capabilities.
Detects outbound HTTPS connections to major generative AI inference endpoints originating from non-browser, non-development tool processes at regular intervals. This activity is indicative of beaconing behavior, where malware (specifically families like LAMEHUG) leverages LLM APIs for dynamic command generation, payload obfuscation, or C2 instruction retrieval, bypassing traditional security controls.
Detects the execution of the sckit Go implant, linked to the MemTensor supply chain compromise, during its host-profiling stage. The rule monitors for the 'sckit' binary being executed with 'stage0' and '--config64' command-line arguments, or being spawned by legitimate runtime interpreters like Node.js or Python, indicating an attempt to profile host information (timezone, language, privileges) and subsequent self-deletion.
Detects anomalous activity from IDEs or AI coding assistant extension host processes (e.g., VS Code, Cursor, JetBrains). The rule alerts when these processes spawn children that access sensitive local credential files, perform network connections to non-standard/unexpected domains, or modify critical CI/CD build script configuration files, which is indicative of a supply chain compromise within the developer environment.
Detects potential runtime polymorphic malware that modifies its own executable or script content while simultaneously communicating with generative AI or LLM API endpoints. This behavior indicates an adversary using LLMs to regenerate obfuscated code or mutate malware signatures dynamically at runtime.
This rule detects potential DNS tunneling activity by identifying DNS requests with suspicious characteristics, such as the use of TXT/NULL query types, high-entropy subdomains, or oversized queries. It further aggregates these suspicious requests per parent domain and flags scenarios where a high volume of requests or unique subdomains are observed, indicating possible data exfiltration or C2 communication.
This rule detects the 'ClickFix' attack pattern, where an adversary tricks a user into copying and pasting a malicious command into a terminal. The command typically involves using 'base64 -D' to decode an attacker-supplied blob directly into a bash interpreter, often initiated by a Terminal emulator and containing strings related to fake reCAPTCHA or robot verification.
Detects the execution of 'setup.js' by the node.js runtime when initiated by 'npm install' or 'postinstall' hooks. This behavior is a common indicator of a malicious supply chain attack where a compromised or malicious npm package executes unauthorized code during the package installation or setup phase.
Detects anomalous behavioral patterns where an AI agent process (e.g., Claude Code, Cursor, Aider) performs multi-category discovery of its own runtime environment, plugins, and host configuration. The rule identifies agents that trigger processes spanning at least three distinct discovery categories (Network, File/Plugin, Software, or System) within a short timeframe, indicating potential reconnaissance of agent-authorized capabilities and tools rather than standard host exploration. Covers T1082, T1518, T1083, T1016
Detects the use of standard command-line utilities (curl, wget) by shell processes (bash, sh) on macOS when fetching a file with a specific force flag. This pattern is indicative of automated attempts to bypass existing file checks or force the download of remote payloads.
TokenGrabber Builder: Nuitka/PyInstaller AV-Evasion Compilation - detects Python builder scripts compiling an embedded stealer.py payload into a native executable via Nuitka or PyInstaller with AV-evasion flags
Detects the modification of AI agent configuration files (e.g., mcp.json, config.json) followed shortly by the agent spawning a child process that matches typical Model Context Protocol (MCP) server execution patterns. This behavior may indicate an attacker has modified the configuration to inject malicious server commands or tools that the AI agent will execute upon next initialization, effectively achieving persistent code execution via the agent's legitimate functionality.
Detects the use of package installation commands (pip, npm, npx) where the package name resembles popular AI/ML libraries (tensorflow, torch, openai, langchain) but is not an exact match for known legitimate versions. This is a common indicator of a typosquatting supply chain attack, where attackers attempt to trick users into installing malicious packages with similar names to trusted software.
Detects instances where AI developer tools or IDE assistants (e.g., Claude, Copilot, Cursor) execute data collection commands (such as directory traversal or archiving) followed by or concurrent with the bulk access of sensitive files (credentials, configuration files) or personal user data.
Detects the modification or creation of a 'loginwindow.plist' file within the '/Library/LaunchAgents/' directory, correlated with a process containing 'renderer' and 'launchctl' or 'LaunchAgents' in its command line. This pattern suggests potential persistence mechanisms being established by a renderer process (e.g., an Electron-based application or browser process).
Detects HTML files that employ FlipBook branding as a lure for password-gated smuggling. The detection identifies client-side JavaScript logic that uses PBKDF2 and AES-GCM to decrypt a hidden redirect URL after user interaction with a password prompt.
Detects AI coding assistants and LLM agent frameworks attempting to access, read, or export sensitive credential files, registry hives, or application secrets. The rule monitors both file system operations on known secret locations and process execution patterns indicative of credential dumping or API token retrieval by these tools.
Detects anomalous behavioral patterns where an AI agent process (e.g., Claude Code, Cursor, Aider) performs multi-category discovery of its own runtime environment, plugins, and host configuration. The rule identifies agents that trigger processes spanning at least three distinct discovery categories (Network, File/Plugin, Software, or System) within a short timeframe, indicating potential reconnaissance of agent-authorized capabilities and tools rather than standard host exploration. Covers T1082, T1518, T1083, T1016
Detects rapid, multi-category system and network enumeration performed by AI-assisted development tools (e.g., Cursor, Claude, ChatGPT). The rule triggers when an AI agent process performs a comprehensive scan of the host environment, specifically covering machine/OS identity, network configuration, process listings, installed software, and access to sensitive credential files within a 10-minute window. It requires at least 5 distinct discovery categories and 6 total events to reduce noise from routine debugging. Covers T1082, T1016, T1057, T1518
Page 16 of 174