Detections
Explore public detection logic contributed by the community across SIEM and rule languages.
7,141 detections
Filters
Last updated
All Time
Detection languages
3,227
1,086
498
476
469
Contributors
1,824
1,607
872
505
300
Categories
2,978
2,471
1,062
1,026
910
Platforms
39,793
7,141
6,515
3,861
3,600
Products / Services
1,325
1,070
970
655
490
MITRE Techniques
2,230
1,319
1,319
875
873
CVEs
66
65
33
32
28
IDS Classtypes
81
79
56
27
10
IDS Protocols
149
98
6
5
4
Detects potential Denial of ML Service (DoMLs) attacks targeting inference endpoints. The rule identifies anomalous behavior characterized by large request payloads (oversized requests) correlated with increased latency or high GPU utilization, suggesting an attempt to exhaust compute resources and impact model availability.
This rule detects potential AI/ML supply chain compromise by monitoring for package installations (via pip, conda, poetry, etc.) from non-standard repositories, direct URLs, or version control systems within data science and CI/CD pipelines (e.g., Jupyter, Airflow, Jenkins). It further correlates this activity with subsequent suspicious outbound network connections from the host, which is indicative of slopsquatting or malicious dependency execution.
Detects recovered scripts/payloads that harvest environment variables, API keys, bearer tokens, and Kubernetes secrets/configmaps into a variable named LOOT
Detects command-line execution patterns that exhibit hallmark traits of LLM-generated code, such as overly verbose and explanatory comments, appearing alongside common obfuscation or decoding primitives (e.g., base64, iex, eval). This combination suggests the use of AI to assist in creating decoy logic or burying malicious payloads within legitimate-appearing scripts to evade detection.
Detects a two-stage activity: first, the potential unauthorized reading or access of LLM provider API keys from files, environment variables, or command-line arguments; and second, the subsequent use of those same credentials to authenticate to LLM provider APIs (OpenAI, Anthropic, Azure, Bedrock) from a different, unrecognized source IP address, indicating potential credential theft and session hijacking.
Detects bulk insertions or content updates into vector databases, document stores, or knowledge bases that contain suspected instruction-override, jailbreak, or role-hijacking phrases, particularly when performed by unauthenticated or unknown identities.
Detects command-line execution patterns that exhibit hallmark traits of LLM-generated code, such as overly verbose and explanatory comments, appearing alongside common obfuscation or decoding primitives (e.g., base64, iex, eval). This combination suggests the use of AI to assist in creating decoy logic or burying malicious payloads within legitimate-appearing scripts to evade detection.
Detects a two-stage activity: first, the potential unauthorized reading or access of LLM provider API keys from files, environment variables, or command-line arguments; and second, the subsequent use of those same credentials to authenticate to LLM provider APIs (OpenAI, Anthropic, Azure, Bedrock) from a different, unrecognized source IP address, indicating potential credential theft and session hijacking.
Detects bulk insertions or content updates into vector databases, document stores, or knowledge bases that contain suspected instruction-override, jailbreak, or role-hijacking phrases, particularly when performed by unauthenticated or unknown identities.
Detects a chained sequence of suspicious activity initiated by agentic AI development tools (e.g., Claude Code, Aider, AutoGen). The rule identifies the execution of these tools followed by local reconnaissance commands, lateral movement attempts, and outbound network connections originating from the same host, indicating potential automated exploitation or credential abuse.
Detects anomalous, chained execution patterns initiated by AI agents integrated with Model Context Protocol (MCP) servers. The rule identifies a multi-stage sequence involving the invocation of browser automation tools followed by shell execution and subsequent outbound network activity, potentially indicating an AI-driven attack chaining filesystem, shell, or cloud-API access in an unauthorized manner.
Detects anomalous, rapid execution of a high volume of diverse offensive security tools (scanners, exploitation frameworks, and C2 agents) from a single parent process within a short window, which is characteristic of automated LLM-driven orchestration or scripted attack tool chaining.
Detects endpoint behavior consistent with AI-assisted exploit development, characterized by the execution of reverse engineering and vulnerability research tools (e.g., IDA Pro, Ghidra, AFL) interleaved with frequent outbound network connections to LLM APIs, followed by the local compilation of a new executable or DLL.
Detects coordinated activity patterns across multiple distinct tenant environments within a one-hour window. The rule looks for shared infrastructure (C2 domains), malicious tool hashes, or specific agentic-CLI command line signatures (e.g., related to orchestration, reconnaissance, or exfiltration) that suggest an AI-orchestrated actor or automated attack campaign affecting multiple organizations simultaneously.
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 bulk insertions or content updates into vector databases, document stores, or knowledge bases that contain suspected instruction-override, jailbreak, or role-hijacking phrases, particularly when performed by unauthenticated or unknown identities.
Detects anomalous, chained execution patterns initiated by AI agents integrated with Model Context Protocol (MCP) servers. The rule identifies a multi-stage sequence involving the invocation of browser automation tools followed by shell execution and subsequent outbound network activity, potentially indicating an AI-driven attack chaining filesystem, shell, or cloud-API access in an unauthorized manner.
Detects anomalous, chained execution patterns initiated by AI agents integrated with Model Context Protocol (MCP) servers. The rule identifies a multi-stage sequence involving the invocation of browser automation tools followed by shell execution and subsequent outbound network activity, potentially indicating an AI-driven attack chaining filesystem, shell, or cloud-API access in an unauthorized manner.
Detects anomalous outbound network connections initiated by security and developer tooling (Trivy, Checkmarx, LiteLLM, Telnyx). This behavior is consistent with supply-chain attacks, such as those attributed to the UNC6780/TeamPCP threat group, where compromised tooling is leveraged to exfiltrate sensitive cloud credentials or API keys to external, non-vendor infrastructure.
Detects anomalous, rapid execution of a high volume of diverse offensive security tools (scanners, exploitation frameworks, and C2 agents) from a single parent process within a short window, which is characteristic of automated LLM-driven orchestration or scripted attack tool chaining.
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.
Page 46 of 341