AI vs Copyright: What Courts Are Saying About Fair Use

AI and Copyright Latest News

• Recent U.S. court rulings — Thomson Reuters vs Ross Intelligence, Bartz vs Anthropic, and Kadrey vs Meta — have clarified that using legitimately acquired texts for training AI models may qualify as “fair use” under copyright law, especially when the output is transformative. 
• However, the legality of using pirated data or the full impact on creative markets remains unresolved, leaving significant legal uncertainty around AI-generated content.

Copyright Challenges in AI-Generated Content

• Generative AI models sometimes replicate or closely resemble original copyrighted works, leading to legal and ethical concerns. 
• The key legal question is whether training AI on such data and producing new content harms the original creator’s market or qualifies as “transformative use.” 
• While some jurisdictions like the U.S. may consider it fair use, others like the EU and U.K. allow it under text and data mining exceptions. 
• However, with AI often trained on large datasets scraped from the internet—including both copyrighted and public domain materials—the global legal consensus on copyright infringement remains unsettled.

Legal Uncertainty Around Databases and Generative AI

• The use of databases and published works for training generative AI models raises complex questions rooted in intellectual property law, contracts, and privacy norms. 
• There is widespread legal ambiguity about whether training AI on IP-protected data and generating outputs from it constitutes infringement. 
• Some countries allow exceptions for fair use, text and data mining, or temporary copying, but these exceptions vary and haven’t been universally tested in court. 
• A major concern is the lack of harmonised global rules, which deepens the uncertainty around who owns IP rights in AI-generated content. 
• This is particularly challenging in copyright law, which traditionally demands human authorship—a standard generative AI does not meet.

US Courts Back Fair Use in AI Training but Warn Against Piracy

• Two significant U.S. court rulings—one favoring Anthropic and the other Meta—affirmed that using copyrighted material to train AI models may qualify as fair use. 
• In Anthropic’s case, Judge ruled that training AI using copyrighted texts is transformative, likening it to how writers learn from existing literature. 
• However, he allowed the trial to proceed on the issue of using pirated data. 
• In Meta’s case, Judge held that the plaintiffs failed to prove that AI outputs diluted the market for original works, thus ruling in Meta’s favor under the fair use doctrine. 
• Both rulings offer legal protection to tech firms for AI training but underscore that piracy remains a legal risk and that mechanisms for compensating copyright holders are necessary.

Implications of AI Copyright Disputes for India

• The ANI vs OpenAI lawsuit is shaping how India’s intellectual property laws apply to generative AI. 
• Under the Copyright Act, 1957, copyright holders have exclusive rights—such as reproduction and adaptation—which require permission unless exempt under Section 52’s fair dealing clause. 
• Although India’s IP laws don’t explicitly mention AI, the prevailing view is that the existing legal framework is adequate to handle AI-related challenges. 
• As a signatory to major global IP treaties, India recognises works by legal entities and provides civil and criminal enforcement, including safeguards against digital rights violations.

Source: TH

‹ ›