AI Fair Use & Copyright: What Creators Need to Know

In a surprising turn, a federal judge ruled that an AI company's use of copyrighted books for training its large language models was 'spectacularly transformative fair use,' even as other courts found similar actions unlawful and major publishers filed lawsuits, Reuters reports. This decision, from Judge William Alsup in Bartz v. Anthropic PBC, granted summary judgment to Anthropic. Some courts are granting summary judgment for AI companies training on copyrighted material, but others and the U.S. Copyright Office reject inherent fair use, finding similar uses unlawful. The legal environment is deeply inconsistent. Therefore, the legal framework for AI and copyright will likely remain fragmented and contested, forcing creators and AI companies to navigate a high-risk, uncertain landscape.

This inconsistency is stark: one court ruled against an AI company, finding its unauthorized use of copyrighted materials for training did not constitute fair use (DGLaw). Simultaneously, The New York Times sued Microsoft and OpenAI, alleging their LLMs infringed copyright by training on its content (ARL). Early, conflicting decisions highlight the profound legal uncertainty surrounding AI's use of copyrighted material, setting the stage for ongoing battles.

Defining Fair Use in the Age of AI

The U.S. Copyright Office rejects the idea that generative AI training inherently qualifies as fair use (The Lawverse). Instead, the Office requires fact-specific, case-by-case analysis, contrasting with some broad judicial interpretations. Meanwhile, the Supreme Court denied certiorari for Dr. Stephen Thaler's appeal on AI-generated artwork, upholding human authorship for copyrightability (Norton Rose Fulbright). This means AI cannot be an author. While the Copyright Office emphasizes a fact-specific approach to fair use, the Supreme Court's stance sets a foundational limit on what AI can claim, even as general principles of fair use for training emerge. AI developers face a dual challenge: they must prove transformative use while navigating the human authorship requirement for their outputs.

When Courts Say 'Fair Use': Wins for AI Developers

In Bartz et al. v. Anthropic, a court ruled AI training on lawfully acquired copyrighted books constitutes fair use (Norton Rose Fulbright). This ruling specifically differentiated between using content for training and storing pirated copies, noting only the former qualifies. A federal judge in Northern California also found training LLMs on lawfully acquired books falls under fair use (The Lawverse). A parallel ruling granted summary judgment to Meta in a similar copyright infringement case (The Lawverse). These rulings suggest a trend: training AI on lawfully acquired material may be fair use if transformative, but pirated data remains off-limits. This encourages AI development, but with a clear line drawn against outright piracy.

The fragmented legal landscape presents significant challenges. For developers, litigation risk remains high despite some favorable judgments (The Lawverse). Relying on broad fair use interpretations could prove precarious if future rulings or legislation align with the U.S. Copyright Office's fact-specific approach. This evolving environment demands constant vigilance and potentially costly legal counsel. Content creators face ongoing uncertainty. While the Supreme Court upholds human authorship (Norton Rose Fulbright), this doesn't automatically prevent AI models from consuming copyrighted human work for training. The judicial embrace of 'spectacularly transformative fair use' in cases like Bartz v. Anthropic is a legal anomaly. It will likely embolden content creators and publishers, like The New York Times and Hachette, to pursue more aggressive legal battles, ensuring a protracted fight over AI's training data (ARL, The Guardian). Individual artists and large media corporations must invest resources in monitoring AI usage and preparing for legal action. Both sides face strategic challenges navigating evolving precedents and potential future legislation.