US judge allows company to train AI using copyrighted literary materials

A US federal judge has issued a mixed ruling regarding Anthropic's AI training practices, deeming the use of copyrighted literary materials for model training as "fair use" due to its transformative nature, a significant win for generative AI developers. However, the ruling also found Anthropic infringed copyrights by storing seven million *pirated* books in a central library, necessitating a December trial for this specific issue. This decision establishes a crucial precedent, supporting the broad application of AI training under fair use while underscoring the critical importance of legitimate data sourcing for AI development.

Analysis

A US federal judge has delivered a pivotal, yet bifurcated, ruling for the generative AI industry, significantly clarifying the legal landscape around training models on copyrighted material. The court found in favor of Anthropic on the core issue of 'fair use', deeming its AI training process "exceedingly transformative" and not a violation of copyright law, which is a major victory for AI developers who rely on vast datasets. This precedent supports the argument that training large language models is a legitimate, creative application rather than mere replication. However, the ruling concurrently exposed a critical operational vulnerability by finding that Anthropic's use of a "central library" containing seven million pirated books constituted copyright infringement. Consequently, while the principle of AI training on copyrighted works was upheld, Anthropic still faces a trial in December for the alleged theft of these materials. This distinction underscores that the legal defensibility of an AI model is contingent not only on the transformative nature of its output but critically on the legitimate sourcing of its training data.

Key Decisions for Investors

• Investors in the AI sector should intensify due diligence on portfolio companies' data sourcing and licensing practices, as the ruling establishes that using pirated source material constitutes a significant and distinct legal risk, even if the training process itself is deemed fair use.
• The 'fair use' victory for Anthropic's training method serves as a de-risking event for the core business model of generative AI companies, potentially driving a valuation premium for firms that can demonstrate clean, legally-sourced data supply chains.
• Closely monitor the outcome of Anthropic's December trial, as the potential damages awarded for using pirated data will set a crucial precedent for the financial liabilities associated with copyright infringement across the AI industry.