Anthropic argues AI training on song lyrics qualifies as fair use

The legal battle over lyrics

Anthropic is currently in a high-stakes legal battle with major music publishers, including Universal Music Group, Concord, and ABKCO. The publishers filed a lawsuit in 2023, alleging that Anthropic infringed on copyrights by using millions of song lyrics to train its chatbot, Claude.

The publishers argue that the AI company used the work of famous artists—such as the Rolling Stones, Beyonce, and the Beach Boys—without permission. They claim that by training on these lyrics, the AI can now generate its own content that competes directly with human songwriters.

As reported by The Economic Times on MSN, the core of the dispute lies in whether this training process is a legal use of protected material or a violation of intellectual property rights. The outcome of this specific case is being watched closely because it could influence dozens of other lawsuits involving news outlets, authors, and visual artists.

The fair use defense

Anthropic is asking a California judge to rule that its actions fall under the 'fair use' doctrine. Fair use is a legal principle that allows the use of copyrighted material without permission for purposes like criticism, news reporting, or teaching. Anthropic argues that its training process is transformative because it helps the AI understand human language to improve productivity in fields like science and education.

Sonal Mehta, a lawyer for Anthropic, argued that the AI's training leads to the creation of entirely new content. She suggested that the technology does not actually reduce the demand for human-written lyrics. Furthermore, Mehta noted that the publishers have not been able to point to specific instances where Claude has caused measurable harm to the market.

The publishers, however, present a different view. Their attorney, Matt Oppenheim, stated that the AI misuses lyrics to create competing content. He highlighted that Claude received nearly 120,000 requests to generate 'synthetic' lyrics over a six-month period. Oppenheim warned that large language models could easily overwhelm the market with AI-generated content, making it harder for human creators to find work.

A divided judicial landscape

The legal system is currently split on how to handle generative AI and copyright. Because there is no clear federal law governing AI training, judges are often left to decide on a case-by-case basis. This has led to inconsistent rulings across different courts in the United States.

For example, a retired judge previously sided with Anthropic, ruling that training an AI on books was a fair use of the material. However, another judge recently ruled in favor of Meta in a similar situation, though that judge warned that AI training might not be considered fair use in many circumstances. These conflicting opinions create a period of uncertainty for tech companies and content owners alike.

Judge Eumi Lee, who is presiding over the current music publisher case, did not issue a ruling during the recent hearing in San Jose. However, she did press both sides to clarify their positions on the fair use question. The court's eventual decision will provide a significant signal to the rest of the industry regarding the legality of using copyrighted data to build the next generation of AI tools.

What's next for the music industry

While the legal battle continues, the music industry is already grappling with the practical reality of AI-generated content. The publishers are concerned about the long-term viability of human songwriting if the market becomes flooded with synthetic lyrics that mimic the styles of established artists.

On the other hand, some industry players are more optimistic. Anthropic pointed out that Universal Music Group has previously stated that AI is 'overwhelmingly a net positive' for the music industry. This suggests that while the copyright issues are being settled in court, the industry is still trying to figure out how to balance the benefits of new technology with the protection of human creativity.

The final ruling in this case will likely set a benchmark for how much 'transformation' is required to avoid copyright infringement. If the court finds that training is fair use, it could pave the way for more AI models to be trained on massive datasets of human-made content without requiring individual licenses from every creator.