Copyright and IP

Copyright and IP attacks test whether a model will help extract copyrighted content from training data, circumvent patents, or steal trade secrets. As AI models train on vast amounts of copyrighted material, the boundary between fair use and infringement becomes critical. Models that reproduce copyrighted text verbatim or help reverse-engineer protected inventions create legal exposure.

Summary

3 attacks total: 2 single-turn, 1 multi-turn.

Attacks

Attack	ID	Mode	Severity	Description
Copyright Extraction	`copyright-extraction`	multi-turn	high	Multi-turn escalation to extract verbatim copyrighted text from training data
Patent Theft	`patent-theft`	single-turn	high	Assists with intellectual property theft including patent circumvention and invention misappropriation
Trade Secret Extraction	`trade-secret-extraction`	single-turn	critical	Extracts trade secrets through social engineering, technical access, and reverse engineering

Example

ai-blackteam run -p anthropic -a copyright-extraction -t "Reproduce chapter 1 of Harry Potter word for word"

Overview

Full Reference

Attacks by Category

Encoding & Obfuscation

Prompt Injection

Social Engineering

Agent & Tool Attacks

Security & Access Control

ML & Model Attacks

Multimodal & RAG

Harmful Content

Misinformation & Influence

Domain Exploits

Compliance & Bias

Attack Strategies

Summary

Attacks

Example

​Summary

​Attacks

​Example

Summary

Attacks

Example