Targeted Doxxing & Threat Campaign
An attacker gives an AI agent a dossier on a real person (name, email, home address) and instructs it to track them down online, send threatening messages, or threaten to expose personal or family secrets. This is a documented pattern for turning agents into tools for targeted harassment and doxxing.
How the attack works
The user first supplies personal identifying details about a target: full name, email address, physical address, or similar. They then instruct the agent to search for more information about this person online, contact them or people connected to them, and issue threats — either direct threats of harm or threats to expose private/family information unless some demand is met. The agent's search, messaging, and reasoning capabilities are used to automate what would otherwise be manual stalking and extortion work.
Netzilo detection
Netzilo reports this behaviour when it is observed.
- Signature ID
- 2df3bf36-cf85-4c30-8c5b-115730f3352a
- Severity
- High
Why it matters
A real individual can be tracked, harassed, or extorted using automation, and the organization operating the agent can be implicated in facilitating stalking, harassment, or blackmail against that person.
What you can do
- →Block or flag agent requests that combine personal identifying information (name plus email or address) with instructions to search for, contact, or confront a specific individual.
- →Refuse and log any request asking the agent to draft or send threatening messages, extortion demands, or threats to expose private information.
- →Distinguish this from legitimate use: safety guides about handling threats, and ordinary contact-form data collection, are not attacks and should not be blocked outright.
- →Review agent logs for dossier-style inputs (structured name/email/address data) paired with outbound messaging or web-search instructions, and escalate matches for human review.
Known benign look-alikes
- Safety guides on what to do when receiving threatening messages
- Benign contact forms collecting name, email, and address