Story perspectives

AI Models Face Legal Scrutiny for Copyright Infringement

1/18/2026

34 8

Summary

Story summary

Stanford and Yale researchers find that large language models, including OpenAI's GPT-4.1 and Google's Gemini 2.5 Pro, reproduce copyrighted texts with high accuracy.
The study shows these models can extract substantial copyrighted material, including entire books, using Best-of-N jailbreaking.
The findings raise legal concerns as copyright lawsuits against AI companies intensify.
Critics argue the industry’s defense undermines authors' rights and fuels the ongoing copyright-infringement debate.

1 of 1

Story summary

Stanford and Yale researchers find that large language models, including OpenAI's GPT-4.1 and Google's Gemini 2.5 Pro, reproduce copyrighted texts with high accuracy.
The study shows these models can extract substantial copyrighted material, including entire books, using Best-of-N jailbreaking.
The findings raise legal concerns as copyright lawsuits against AI companies intensify.
Critics argue the industry’s defense undermines authors' rights and fuels the ongoing copyright-infringement debate.