Brandon Rodriguez
@BMRodriguezTrusted contributor
3 followers22 downloads772 copies15 likes20,508 views
123 detections
Filters
Last updated
All Time
Detection languages
123
Categories
43
30
18
15
13
Platforms
84
23
10
7
7
Products / Services
39
9
7
7
4
MITRE Techniques
36
36
28
27
21
Detects the VS Code CLI (code-insiders / code) being used to install a remote tunnel as a service, run from outside normal application install locations (e.g. the Windows debug folder). Keyed on the tunnel service install behavior and unusual location rather than a specific file path, so renamed or relocated copies still match. Abuse enables covert remote access via Microsoft tunnel infrastructure. Rotation test: still fires if the actor changes the directory, flags order or infrastructure, as long as the binary is run with tunnel + service install outside standard paths.
Detects instances where Microsoft Copilot memory content (identified by hash) remains identical before and after a significant account security event, such as a password reset or session revocation. This behavior suggests that an adversary may have injected malicious or persistent data into the user's Copilot memory, which survives standard account remediation actions.
Detects Microsoft Copilot assistant responses that explicitly claim a documented CVE is safe or poses no risk. This behavior is consistent with potential memory-driven prompt manipulation or disinformation, where an attacker has poisoned the assistant's memory to downplay known security vulnerabilities.
Detects potential indirect prompt injection attacks against Microsoft Copilot where a 'summarize' action targeting a webpage is immediately followed by suspicious assistant behavior changes such as memory updates, suggesting the webpage content contained malicious instructions.
Detects instances where Microsoft Copilot memory content (identified by hash) remains identical before and after a significant account security event, such as a password reset or session revocation. This behavior suggests that an adversary may have injected malicious or persistent data into the user's Copilot memory, which survives standard account remediation actions.
Detects Microsoft Copilot interaction URLs containing the 'autorun=1' parameter alongside a 'q=' prompt parameter. This combination indicates a potential attempt to execute a prompt in the user's Copilot session without explicit user confirmation upon link click, a technique sometimes referred to as a CoSnitch or one-click attack.
Detects Microsoft Copilot interaction URLs containing the 'autorun=1' parameter alongside a 'q=' prompt parameter. This combination indicates a potential attempt to execute a prompt in the user's Copilot session without explicit user confirmation upon link click, a technique sometimes referred to as a CoSnitch or one-click attack.
Detects Microsoft Copilot assistant responses that explicitly claim a documented CVE is safe or poses no risk. This behavior is consistent with potential memory-driven prompt manipulation or disinformation, where an attacker has poisoned the assistant's memory to downplay known security vulnerabilities.
KQL Query
KQL Query
Page 1 of 13