Back to all articles
•Technology

AI Training Using Copyrighted Material Poses Legal Challenges

View original source

AI models like ChatGPT, Gemini, and Claude are trained on vast collections of databases, including copyrighted works, often without the authors’ knowledge or consent. This practice raises complex legal issues regarding copyright infringement.

  • Cathy Gellis, an attorney in intellectual property, explains that the situation is complicated and lacks straightforward legality.
  • A landmark ruling by Judge William Alsup fined Anthropic $1.5 billion for using pirated books to train models. Yet, Alsup ruled that the practice of AI training on these works was legal, comparing it to a writer's study of literature.
  • Gellis suggests this ruling favors AI firms and highlights that copyright laws, rooted in the 1976 Copyright Act, are outdated for modern AI advancements.
  • Jason Henderson, a Senior Attorney, emphasizes the uncertainty in legal interpretations regarding AI and copyrighted material, especially concerning fair use.
    • Fair Use laws determine whether use of copyrighted work is transformative enough to be permissible, considering factors like the purpose of usage and market impact.
    • A cited case involved Thomson Reuters suing Ross Intelligence for training on its content to make a competing platform, ruled as not fair use by Judge Stephanos Bibas.
  • Ryan Gellis elucidates the distinction between AI training on copyrighted works and AI-generated content, citing Thaler v. Perlmutter, which ruled purely AI-generated works as non-copyrightable.
  • Pending litigation continues to create uncertainty over legal frameworks, with initial court decisions influencing future cases. Gellis stresses the importance for AI companies to heed ongoing legal developments.