Three years into the AI copyright fight, nobody has settled whether training on copyrighted work is legal. But the cases point somewhere more specific. Courts are asking two narrower questions: where the material came from, and whether anyone can prove the output caused harm.
