Goksel Atakan
@gokselatakanIstanbul, Turkiye GoXDR.fyiTrusted contributorCompletionist
0 followers73 downloads1,580 copies38 likes3,484 views
91 detections
Filters
Last updated
All Time
Detection languages
81
10
Categories
18
16
16
15
12
Platforms
34
22
18
16
13
Products / Services
18
14
12
11
11
MITRE Techniques
42
28
21
20
18
CVEs
3
2
2
1
1
This rule identifies devices with a high concentration (5 or more) of local AI agents reported within a 30-day period. This could indicate potentially unauthorized mass deployment or misconfiguration of AI agent software on specific endpoints.
AD Group Changes and Membership Activity
Cortex XDR
This rule monitors Active Directory security event logs for modifications to local, global, and universal groups. It specifically tracks events related to the creation, deletion, and membership changes of AD groups, providing visibility into potential privilege escalation or unauthorized administrative activity within the domain.
This rule identifies accounts (by SID) that have registered multiple local AI agent instances. It aggregates agent registration information over a 30-day period, summarizing the total count of agents and listing specific device names and IDs associated with each account to identify potential anomalous behavior, such as credential abuse or unauthorized mass deployment of local AI tools.
This rule extracts and audits inventory information from local agents, including vendor, device name, account context, and associated process information. It is designed to provide visibility into the state and configuration of installed local agents within the environment.
This rule analyzes recent agent telemetry to identify and summarize the distribution of local agent vendors present in the environment over a 30-day lookback period. It extracts vendor information from agent metadata and provides a frequency count to assist in auditing deployed infrastructure agents.
Device Name Search
Cortex XDR
This rule performs a telemetry data retrieval query for a specific agent hostname. It outputs detailed fields related to network connections and process executions (including command lines, user context, and process image names) from the XDR dataset for auditing or investigative purposes.
This takes the earliest available snapshot per agent instead of comparing two states. If an agent is born already carrying MCP access and org-wide sharing there is no baseline to drift from, so this needs its own check rather than a diff.
This rule detects potentially malicious modifications to agent configurations by monitoring for multiple significant changes within a short time frame (2 days) compared to a 14-day baseline. Monitored indicators include: changes to instructions, addition of new Model Context Protocol (MCP) servers, addition of new owners, and expansion of access permissions to include the entire organization. An alert is triggered when an agent exhibits at least two of these signals simultaneously.
The query builds a 14-day baseline snapshot and a 2-day current snapshot per agent from AgentsInfo, computes the four signals independently, then correlates them by AgentId and surfaces only agents with 2 or more signals firing in the same window. Output is a prioritized triage list instead of four separate alert streams.
The query builds a 14-day baseline snapshot and a 2-day current snapshot per agent from AgentsInfo, computes the four signals independently, then correlates them by AgentId and surfaces only agents with 2 or more signals firing in the same window. Output is a prioritized triage list instead of four separate alert streams.
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 detects anomalous or risky sign-in events for user accounts that have recently configured an active out-of-office (OOF) auto-reply. An attacker may leverage a user's known absence to gain access to their email or other corporate resources using compromised credentials, as the user is less likely to notice suspicious account activity while away.
Page 9 of 10