Public crash investigation records exist because aviation safety depends on transparency. But the same cockpit voice recordings that help investigators reconstruct what went wrong in a fatal accident now provide enough raw audio for AI tools to rebuild the voices of pilots who didn't survive.
That's what happened. AI users pulled documents from a public NTSB (National Transportation Safety Board) docket and used them to recreate voices of deceased pilots. The agency responded by restricting public access to those materials - a significant step for an organization whose entire credibility rests on openness.
Voice cloning has become far more accessible in the past two years. Tools that once required hours of audio and technical expertise now need seconds of sample material to produce convincing output. A crash docket contains exactly what these tools need: cockpit voice recordings, recorded pilot interviews, air traffic control communications. The raw material was always there. The barrier to using it just disappeared.
The NTSB hasn't detailed what exactly was done with the voice recreations, but considered it serious enough to pull a public docket. That's the tell.
The harder question is where this ends. Crash data isn't the only public record containing usable voice samples. Court recordings, deposition transcripts, congressional testimony, recorded earnings calls - voice-clonable audio is embedded throughout the public record. The pilots in these dockets could not consent to having their voices reconstructed. No current law prevents it.
The technical barrier to cloning a voice from public records is essentially zero now. The ethical and legal frameworks for whether you should have barely been written, and the NTSB pulling a docket is what it looks like when that gap stops being theoretical.