Determining what actually escaped the studio vault requires separating finished master files from legacy archival scraps. Much of what circulated during the frenzy was not pristine album audio. Instead, bad actors repackaged abandoned demo recordings from the When We All Fall Asleep, Where Do We Go? sessions, attaching current song titles to vintage, unreleased tracks.
Superstar artists continuously write material that never makes an official tracklist. Hard drives contain hundreds of acoustic voice memos, alternate vocal stems, and abandoned instrumentals. When leak opportunists smell blood, they flood online forums with these orphaned files to harvest clicks and Discord subscriptions. Audio engineers who analyzed the waveform data of the pre-release leaks noted drastic variance: snippets of authentic album tracks sat at heavily compressed bitrates, likely captured through illicit phone recordings inside vinyl manufacturing plants, while older unreleased tracks arrived in clean stereo mixes.
The confusion grew when AI voice models entered the mix. By training neural networks on Eilish's distinct whispering delivery, bad actors synthesized dozens of fake snippets. This audio pollution made verifying genuine leaks an arduous task for casual listeners, inadvertently blunting the impact of the authentic leaks.
