In a startling demonstration of AI's rapid capabilities, a user on X claimed it took just 10 minutes using OpenAI's Codex to reconstruct the last words of deceased pilots from a publicly released spectrogram. The incident highlighted the growing ease with which AI can bring back voices of deceased people, including those involved in fatal accidents. The ability to resurrect voices from even minimal data sets like spectrograms signals a profound challenge to established norms.
The NTSB’s mission is to provide public data for safety analysis, but that very data is now being used by AI to recreate the voices of the deceased, causing distress and forcing data restrictions. The US National Transportation Safety Board (NTSB) temporarily suspended public access to its database of civil transportation accidents (Benzatine Infotech, Ars Technica).
Agencies will likely implement stricter controls or redactions on publicly available incident data, trading full transparency for ethical protection, which could inadvertently hinder independent safety analysis. The NTSB pulled its docket system offline after people used AI to recreate the voices of pilots killed in a plane crash (Engadget), with these re-created voices circulating online (TechCrunch).
How AI Recreated the Voices
AI was used to recreate the last 30 seconds of pilots' conversation from a spectrogram released by the NTSB, according to Engadget. One user on X specifically stated it took them just 10 minutes using OpenAI's Codex to reconstruct audio from the NTSB-released spectrogram.
Beyond audio files, an AI-generated rendering depicting cockpit audio of UPS flight 2976 was created and circulated using only written documents submitted during a recent hearing, as reported by Mashable. The capability to reconstruct voices from both visual data like spectrograms and textual records reveals a new frontier of digital ethics.
