The labs' memorization defense was always a bet on technical opacity. Courts cannot subpoena model weights and read them; plaintiffs cannot directly demonstrate that a billion-parameter model contains a copy of a specific book. That opacity has functioned as a structural advantage in every major copyright case filed since 2023. The 'Alignment Whack-a-Mole' paper eliminates it — not by reading the weights, but by demonstrating that the content they contain can be extracted through a commercially available process. When the SSRN account posted the finding that finetuning 'can unlock 85–90%…