There’s no agreement on whether training a large language model on copyrighted content is legal. That’s not a gap waiting to be filled, it’s the active state of the law in 2026, on both sides of the Atlantic.
Three distinct tracks are running in parallel right now. They don’t coordinate with each other. They may produce outcomes that directly conflict. Understanding all three isn’t optional for any AI company that trains models, uses third-party models trained on scraped data, or operates in markets where this question is moving toward resolution.