AI copyright news, graded by source

Suno's answer admits its training data includes essentially all reasonable-quality music files accessible on the open internet, and stakes its defense on fair use

Answering the record labels' June 24, 2024 complaint, Suno said on Aug. 1, 2024 that its model was built by showing a neural network tens of millions of recordings, essentially as many as can be located, many of which the labels control, so that its training data 'includes essentially all music files of reasonable quality' accessible on the open internet, while respecting paywalls and password protection. It said that whether UMG's particular version of Johnny B. Goode was in the data, many UMG recordings probably were. Suno argued that making intermediate copies to build a product whose outputs do not infringe is fair use, pointing to Google Books, Perfect 10 and the Sega and Oracle decisions, and that the labels' concession that they were not alleging the outputs themselves infringe would prove fatal to their claims. It also described guardrails that block prompts naming particular artists and audio-upload checks, and accused the labels of misusing the tool to generate the examples in their complaint. The labels later learned in discovery that the audio came from YouTube by stream-ripping, which became the subject of a DMCA claim added in 2026.