Comment by richardfey

They could use an agent to summarise the source material, and then train models on those summaries, and claim that some sort of clean-room training has happened?