Detections
Explore public detection logic contributed by the community across SIEM and rule languages.
46 detections
Filters
Last updated
All Time
Detection languages
22
6
6
6
2
Contributors
7
7
6
5
4
Categories
31
17
9
7
5
Platforms
15
13
8
7
7
Products / Services
9
7
6
6
5
MITRE Techniques
18,108
15,614
12,573
8,384
6,204
CVEs
2
1
1
IDS Classtypes
5
1
IDS Protocols
5
1
Detects browser-side malicious activity involving communication with Cloudflare Workers (workers.dev) endpoints typically associated with C2 beaconing (relays, ICE, socket) or the retrieval of trojanized JavaScript dependencies like 'jquery.mousewheel' which often precede second-stage agent execution.
This rule detects attempts to jailbreak Large Language Models (LLMs) by monitoring event logs and process command lines for indicators of 'Prompt Shields' activity. It specifically identifies when a jailbreak attempt is either detected or blocked by security controls.
This rule detects potential prompt injection attempts targeting AI systems by looking for specific keywords related to AI models (gpt, claude, copilot) in HTTP logs, combined with suspicious characters or commands commonly used in injection attacks (e.g., <script>, javascript:, eval(). The rule triggers if 5 or more such attempts are observed from a single IP within an hour.
Detects malicious Parameter-to-Prompt (P2P) injection attempts targeting Microsoft 365 Copilot by monitoring search query parameters for indicators of natural-language command injection or HTML-based payload injection (e.g., img tags, source manipulation, or data extraction commands).
This rule detects attempts to perform Large Language Model (LLM) prompt injection via HTTP requests. It monitors proxy traffic for common jailbreaking or adversarial instruction patterns such as 'ignore previous instructions', 'bypass restrictions', or 'fictional environment' within URI parameters or request bodies. Such patterns are indicative of an attacker attempting to manipulate the behavior of an AI/LLM application.
This rule monitors for security alerts related to Azure AI services, specifically detecting credential theft attempts and LLM jailbreak attempts that have been blocked or detected by Azure's built-in content filtering mechanisms.
This rule monitors for security alerts related to Azure AI services, specifically detecting credential theft attempts and LLM jailbreak attempts that have been blocked or detected by Azure's built-in content filtering mechanisms.
Detects attempts to bypass URL validation and access sensitive GCP metadata endpoints (169.254.169.254 or metadata.google.internal) using a double-slash (//) prefix injection pattern within URL parameters. This technique is often used in Server-Side Request Forgery (SSRF) attacks to read local resources or metadata credentials from Cloud Run or other GCP compute services.
Detects network requests to Microsoft 365 Copilot Enterprise Search containing suspicious HTML tags or prompt injection keywords in the query parameter, indicating a potential Parameter-to-Prompt (P2P) injection attempt (CVE-2026-42824). Specifically looks for HTML image tags (<img src=), the variable '$TITLE', or the Bing SSRF endpoint 'bing.com/images/searchbyimage' within the URL-decoded 'q' parameter.
This rule detects potentially malicious search queries within Microsoft 365 network traffic. It monitors HTTP headers and request bodies for search parameters that contain common web injection markers, such as script, img, or svg tags, and other suspicious keywords indicative of attempted Cross-Site Scripting (XSS) or search-based exploitation.
Detects Log4Shell JNDI injection attempts
This rule detects attempts to perform Large Language Model (LLM) prompt injection via HTTP requests. It monitors proxy traffic for common jailbreaking or adversarial instruction patterns such as 'ignore previous instructions', 'bypass restrictions', or 'fictional environment' within URI parameters or request bodies. Such patterns are indicative of an attacker attempting to manipulate the behavior of an AI/LLM application.
Detects network requests to Microsoft 365 Copilot Enterprise Search containing suspicious HTML tags or prompt injection keywords in the query parameter, indicating a potential Parameter-to-Prompt (P2P) injection attempt (CVE-2026-42824).
This rule detects Content Security Policy (CSP) violations related to 'netdna-ssl.com' or 'wpengine.netdna-ssl.com' domains. Specifically, it looks for violations where the blocked URI contains 'netdna-ssl.com' or the violated directive is 'script-src', 'font-src', or 'connect-src'. This could indicate attempts to inject malicious content or bypass CSP protections.
This rule detects potential prompt injection attempts targeting AI systems by looking for specific keywords related to AI models (gpt, claude, copilot) in HTTP logs, combined with suspicious characters or commands commonly used in injection attacks (e.g., <script>, javascript:, eval(). The rule triggers if 5 or more such attempts are observed from a single IP within an hour.
This rule detects potential prompt injection attempts targeting AI systems by looking for specific keywords related to AI models (gpt, claude, copilot) in HTTP logs, combined with suspicious characters or commands commonly used in injection attacks (e.g., <script>, javascript:, eval(). The rule triggers if 5 or more such attempts are observed from a single IP within an hour.
Detects injection of malicious scripts into legitimate websites for watering hole attacks
Detects hidden iframes used for drive-by download attacks
This rule detects the presence of the string 'document.write(unescape(' in HTTP response bodies. This pattern is commonly used in web-based attacks to inject malicious content, often obfuscated, into legitimate web pages served to clients. The 'flow:to_client,established' indicates that the rule is looking at traffic flowing from the server to the client over an established connection, specifically within the HTTP data payload ('file.data'). The 'nocase' option ensures the detection is case-insensitive.
This rule detects potential malicious script injection from ad networks. It looks for HTTP responses (flow:to_client) that contain common ad-related terms in the HTTP referer header (ad., ads., advert., banner.) and also contain JavaScript code obfuscation techniques like 'eval(' and 'fromCharCode' within the file data. This combination suggests an ad network might be serving malicious JavaScript to clients.
This rule detects HTTP response content injection via a Man-in-the-Middle (MitM) attack. It specifically looks for HTML content (indicated by 'Content-Type: text/html') in the HTTP response that contains a '<script' tag and a 'src' attribute pointing to an IP address (e.g., 'src="https?://[0-9]{1,3}\.[0-9]{1,3}'). This pattern suggests that an attacker might be injecting malicious scripts into legitimate web traffic.
Page 1 of 3