Offensive Security Tool Reference in Agent Content
This rule flags when an AI agent's actions - the commands it runs, tool calls it makes, or files it writes - mention well-known offensive security tools like exploit frameworks, network scanners, password crackers, or phishing kits. It's a sign the agent has been directed (or manipulated) into performing hacking activity rather than its intended task.
How the attack works
An attacker prompts, tricks, or otherwise directs an AI agent into using or referencing tools such as exploit frameworks, credential dumpers, network scanners, privilege escalation utilities, or phishing kits. The agent then executes a command, makes a tool call, or writes a file containing the name or signature of one of these tools. The rule matches on that reference appearing in the agent's activity, regardless of how the agent was steered into producing it.
Netzilo detection
Netzilo can block this behaviour when it is observed.
- Signature ID
- a0d1e2f3-a4b5-4c67-8901-d2e3f4a5b6c7
- Severity
- High
Why it matters
If unnoticed, this activity can lead to credential theft, network reconnaissance, privilege escalation, or successful phishing - using the agent's own execution environment and permissions as the launch point.
What you can do
- →Review any flagged agent session to confirm whether the tool reference was part of an authorized security test or training exercise.
- →Restrict agent environments from installing or executing arbitrary binaries and scripts, especially ones matching known hacktool names.
- →Require human approval for agent actions that involve running system commands, downloading executables, or writing scripts to disk.
- →Maintain an allowlist of expected agent tools and commands, and treat any deviation as worth investigating.
Known benign look-alikes
- Authorized penetration testers using these tools in a sanctioned engagement
- Security training platforms running CTF challenges