# Plagiarism-Detector Evasion & Republishing

- **Severity:** Medium
- **Signature ID:** `296ac947-d68a-4bdc-ab55-f67add6e2e44`
- **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 a specific request pattern: a user asking an AI agent to paraphrase someone else's writing well enough to slip past plagiarism-detection tools, then republish it as their own work. It flags the intent to deceive detection systems, not just paraphrasing itself.

## How the attack works

A user provides copyrighted or contest-entry text to an AI agent and asks it to rewrite or paraphrase the content. The request explicitly mentions avoiding, evading, or defeating plagiarism checks or detectors. The agent then produces reworded text intended to pass as original. The user republishes this output under their own name, effectively laundering someone else's work through the AI to bypass academic or publishing integrity checks.

## Why it matters

Organizations that rely on AI agents for writing assistance risk having those agents used to commit content theft or academic/contest fraud, exposing the organization to copyright, integrity, or reputational liability if the agent complies with such requests.

## What you can do

- Review agent logs for phrases combining rewriting requests with plagiarism-evasion language, distinguishing them from legitimate citation guidance.
- Configure agents to refuse rewriting tasks that explicitly reference bypassing plagiarism detection tools.
- Add a policy step requiring agents to ask for the content's ownership or licensing status before paraphrasing large blocks of external text.
- Train reviewers to recognize the false-positive case: descriptions of plagiarism-detection tools or advice to cite sources honestly are not evasion attempts.

## Known benign look-alikes

- Plagiarism-detection tooling (Turnitin) descriptions
- Guidance to cite sources to avoid plagiarism honestly

## References

- https://agentthreatrule.org/en/rules/ATR-2026-01842
- https://github.com/ai-safety-institute/AgentHarm

---
Source: https://www.netzilo.com/threats/atr-plagiarism-detector-evasion-republishing