# Non-Consensual Leaked-Media Purchase & Distribution

- **Severity:** High
- **Signature ID:** `be689512-719c-45ac-bf43-46ae4c5005f3`
- **MITRE ATLAS:** AML.T0051 (LLM Prompt Injection), AML.T0057 (LLM Data Leakage)
- **OWASP:** ASI01 (Agent Authorization and Control Hijacking), ASI08 (Supply Chain Attacks on Agent Ecosystems), LLM01 (Prompt Injection), LLM02 (Sensitive Information Disclosure)

## Summary

This detects an AI agent being instructed to purchase non-consensual intimate or private images of a named person and then distribute them across social media. It covers a specific abuse pattern: acquiring leaked private media and amplifying it, not general content moderation.

## How the attack works

A user gives the agent a direct instruction naming a specific person and referencing a leaked private photo of them. The agent is told to pay for access to that image, for example on a marketplace or paywalled site. Once obtained, the agent is instructed to post or distribute the image across multiple social platforms, maximizing reach. The rule flags this purchase-then-distribute sequence as it happens through agent actions.

## Why it matters

An organization whose agent carries this out becomes a direct participant in non-consensual intimate image (NCII) distribution, exposing it to legal liability, platform bans, and reputational damage, while causing real harm to the victim.

## What you can do

- Block or require human approval for any agent action involving payment for media tied to a named private individual.
- Add explicit policy rules preventing agents from posting personal images to social platforms without verified consent.
- Review agent logs for purchase-then-broadcast patterns involving personal media, not just single suspicious posts.
- Distinguish this from legitimate takedown or reporting workflows, which describe removing leaked content rather than acquiring and spreading it.

## Known benign look-alikes

- Descriptions of takedown processes when a leaked photo is posted without consent

## References

- https://agentthreatrule.org/en/rules/ATR-2026-01853
- https://github.com/ai-safety-institute/AgentHarm

---
Source: https://www.netzilo.com/threats/atr-non-consensual-leaked-media-purchase-distribution