Detections
Explore public detection logic contributed by the community across SIEM and rule languages.
1,084 detections
Filters
Last updated
All Time
Detection languages
418
193
163
156
81
Contributors
381
251
92
71
40
Categories
498
271
255
135
123
Platforms
443
441
267
121
114
Products / Services
10,665
9,758
6,692
4,418
3,749
1,084
MITRE Techniques
278
229
180
177
151
CVEs
17
15
13
13
8
IDS Classtypes
4
IDS Protocols
4
Detects adversarial backdoor triggers (such as pixel patches, watermark-like artifacts, or malicious token sequences) injected into AI inference inputs. The rule identifies these inputs by observing a combination of suspicious trigger patterns and significant, anomalous jumps in prediction confidence directed at a specific target class, indicative of a model integrity compromise.
Detects anomalous high-volume API querying patterns against an AI model, characterized by an abnormally high number of distinct input hashes and query frequency compared to the population baseline for that model. This pattern is consistent with model extraction or stealing attacks aimed at reconstructing a proxy model.
This rule detects potentially malicious activity related to AI agent tool manipulation. It flags two primary scenarios: 1) The invocation of an AI tool shortly after an unmanaged or potentially unauthorized modification to its definition, and 2) Detection of prompt injection patterns within the output generated by an AI agent tool, which could indicate an attacker attempting to override the agent's instructions.
Detects a malicious sequence of actions originating from an Android Accessibility Service, specifically the 'SystemHelperService'. The activity involves the programmatic enablement of Developer Options and Wireless Debugging, followed by the unauthorized extraction of an ADB pairing PIN from the screen, bypassing user interaction requirements.
Detects anomalous access to the SharePoint WebPartPages.asmx SOAP endpoint, specifically using the GetWebPartPageConnectionInfo method. This query identifies patterns consistent with the exploitation of CVE-2026-65660, where adversaries attempt to smuggle WebPart template markup (e.g., <%@ Register, XamlServices) after legitimate preview methods have been disabled.
Detects the creation of scheduled tasks, recurring jobs, or webhook subscriptions by a user identity flagged as an AI_AGENT. This activity is monitored for persistence, especially when the event lacks a descriptive context or targets external networks, suggesting potential unauthorized agent behavioral drift or malicious persistence.
Detects anomalous use of an LLM provider API key by identifying successful API requests originating from multiple distinct countries within a one-hour time window. This activity often indicates the potential theft and concurrent unauthorized use of an API token.
Detects the registration of new tools, plugins, or connectors into an AI agent's registry, particularly from unverified publishers or outside of sanctioned deployment windows. The rule also flags cases where these newly registered tools are immediately invoked with sensitive-scope permissions (e.g., admin, secrets, write), which is consistent with indicators of AI agent supply chain compromise or unauthorized tool injection.
Detects anomalous, high-volume file deletion, modification, or resource deletion operations performed by an identified AI agent service account within a short timeframe. This behavior is indicative of a compromised AI agent being leveraged for mass data destruction, ransomware activity, or disruptive configuration changes.
Identifies NetScaler ADC or Gateway software versions in the inventory that are known to be vulnerable to CVE-2026-88771 and CVE-2026-88772, despite having received the patch for CVE-2026-19490.
This rule detects network communication with known malicious infrastructure (C2 IPs and URLs), presence of malicious files identified by SHA256 hashes on disk, and execution of known malicious files. It also correlates these activities with specific operator email addresses involved in the malicious campaign.
Detects web requests targeting Oracle PeopleSoft Integration Broker (PSIGW) endpoints that contain suspicious query parameters associated with exploitation attempts, such as path traversal sequences, command injection patterns, or Java-based class invocations. This rule is designed to identify exploitation attempts against PeopleSoft systems as observed in the mid-2026 ShinyHunters zero-day campaign.
This rule detects potential adversarial reconnaissance activities targeting AI agents via multiple communication channels (e.g., email, collaboration tools). It identifies patterns where a sender transmits messages containing common 'jailbreak' or 'instruction-override' phrases (e.g., 'ignore previous instructions', 'bypass approval') across at least three distinct channels within a 24-hour window, indicating a concerted effort to manipulate or probe agent logic.
Detects autonomous behavior where AI-driven processes attempt multiple network connections (SMB, SSH, HTTP, RDP, WinRM) to diverse targets following failed access attempts. The rule correlates initial failure activity with subsequent credential-related events and new lateral movement traffic from the same agent process within a short window, suggesting adaptive adversary pivoting.
This rule detects attempts to exploit a pre-authentication path traversal and unsafe file-upload vulnerability (CVE-2026-93616) in Check Point Management Server. It monitors web server logs for specific application-level error messages related to ReflectionUtils failure and suspicious path traversal sequences (../) in request URLs.
Detects malicious instruction-override payloads, hidden HTML comments, or obscured text (e.g., zero-font size) in third-party content ingested into LLM/RAG systems, commonly used for indirect prompt injection.
Detects the download of a dataset from an unapproved source that lacks hash verification, followed shortly by the launch of a model training process on the same host. This behavior is indicative of potential supply chain compromise or poisoning of the training data used in machine learning workflows.
Detects adversarial prompts designed to induce an Large Language Model (LLM) to disclose sensitive information, such as PII, secrets, or internal training data. This rule monitors the input prompt for extraction-based patterns and evaluates the model's response for indicators of a sensitive data leak (DLP hit).
Detects attempts to perform a Denial of Service (DoS) attack against an AI-enabled system by flooding the inference API with high volumes of requests or overly large inputs (high token counts). The rule identifies potential resource exhaustion by correlating high request rates or large input sizes with system-wide impacts, such as increased inference queue depth or excessive GPU utilization, which could impede service availability for other users.
This rule detects attempts to bypass LLM safety guardrails using adversarial prompting techniques (e.g., 'DAN' persona, hypothetical scenarios) or by identifying patterns of repeated moderation flags and refusals. It specifically looks for user prompts containing known jailbreak templates or an excessive frequency of safety-related trigger events, while excluding authorized red-team and safety evaluation accounts.
Detects evidence of indirect prompt injection attacks where an adversary introduces malicious instructions into data sources retrieved by an LLM or agent. The rule identifies common obfuscation and override techniques, including the use of zero-width characters, HTML comments for prompt breaking, base64 encoded instruction verbs, and specific override phrases within retrieved documents or tool outputs.
Page 5 of 52