The Atlantic made the music AI stole searchable

Atlantic reporter Alex Reisner dug into the invisible pile of music feeding AI models and built a searchable database of the tracks. He found four datasets — two enormous ones with 12 million and 9 million songs, plus two smaller ones with over 100,000 each. Google and Stability have used them. The Free Music Archive set is free for personal listening but not for training AI.

The numbers are big, but what's bigger is the question. These datasets are being used to train models that now generate music, and the music industry is starting to push back. The database gives artists a way to find their work and challenge how it's being used.

Why this matters for us: When AI trains on music without asking, it's the same old story — the work of working musicians powers the tech while the tech gets the credit. Now that the Atlantic made it searchable, artists can fight back.