CopyrightCatcher, from Patronus AI, detects when large language models reproduce copyrighted content verbatim — aimed at organisations that need to manage infringement risk in AI outputs.
Benefits & outcomes
In Patronus AI's adversarial test of how often models reproduced copyrighted material: GPT-4 on 44% of prompts, Mistral Mixtral-8x7B on 22%, Llama-2-70b-chat on 10%, Claude 2.1 on 8%.
Notes & quotes
"LLM training data often contains copyrighted works, and it is pretty easy to get an LLM to generate exact reproductions from these texts."
"It is critical to catch these reproductions, since they pose significant legal and reputational risks for companies that build and use LLMs in production systems."
Publisher / company: Patronus AI
Date of mention: 06/03/2025
Source: https://www.patronus.ai/blog/introducing-copyright-catcher