GENERATIVE AI
FAIR USE OR FOUL PLAY? THE LEGAL KNOT NO ONE CAN UNTANGLE
active users to 700 million and valuing the company at US $ 500bn. Yet critics deemed it“ overdue, overhyped and underwhelming.”
Meanwhile, competitors caught up with formidable releases of their own. Google DeepMind’ s Gemini 2.5 topped the LMArena leaderboard in March. Anthropic’ s Claude 4 achieved a record 72.5 % success rate on software engineering benchmarks in May and Meta’ s Llama 4 family set a new benchmark with an unprecedented 10 million-token context window in April.
2025’ s most perplexing legal saga involved two nearly identical cases with opposite outcomes. When Anthropic faced copyright claims, a federal judge ruled that training AI models on copyrighted books constituted fair use: a transformative use that benefits society. However, the court drew a line at using pirated copies, holding Anthropic liable for accessing unauthorised materials.
Days later, a different California judge hearing Meta’ s case reached a startlingly different conclusion: not only was training on copyrighted material fair use, but obtaining those materials through piracy was also excused given the difficulty of licensing at scale.
These contradictory rulings expose fundamental uncertainty in applying decades-old copyright doctrine to cutting-edge AI. Key unresolved questions include: Does the commercial nature of AI companies matter? Is the amount of copyrighted material used relevant? Can artists opt out? Until the Supreme Court weighs in – likely years away – AI companies operate in legal limbo, with every training run carrying potential billion-dollar liability. aimagazine. com 169