US transportation regulator NTSB pulled its accident reviews after the audio recreations have been uploaded on-line.
The Nationwide Transportation Security Board (NTSB) has pulled its docket system offline after folks used info uploaded to it to recreate the voices of pilots killed in a airplane crash with AI. As CNN reviews, the company lately uploaded information full of particulars concerning the November 4, 2025 crash involving UPS flight 2976. One of many airplane’s engines separated from the wing throughout takeoff from Louisville, Kentucky, killing three crew members and 12 folks on the bottom.
Whereas the NTSB uploads accident reviews that the general public can entry, it isn’t allowed by federal regulation to launch cockpit audio recordings “as a result of extremely delicate nature of verbal communications contained in the cockpit.” The NTSB’s uploads for the UPS crash included 1000’s of pages of reviews and a video displaying the engine’s separation. In addition they included a transcript of the black field recordings and a PDF file with a spectrogram, which reveals a graphic illustration of the recorded audio within the cockpit. It is by that spectrogram that folks have been in a position recreate the final 30 seconds of the flight, whereas the pilots have been struggling to disable the airplane, utilizing synthetic intelligence.
“The NTSB is conscious that advances in picture recognition and computational strategies have enabled people to reconstruct approximations of cockpit voice recorder audio from sound spectrum imagery launched as a part of NTSB investigations, together with the continued investigation of the crash final yr of UPS flight 2976 in Louisville, Kentucky,” the board stated in its announcement. “The NTSB docket system is briefly unavailable as we look at the scope of the problem and consider options.”
Whereas know-how to show spectrograms again into audio has existed for some time now, AI has made it simpler to take action by anyone who has entry to it. As Ars Technica reviews, one consumer on X stated it took them 10 minutes utilizing OpenAI’s Codex to reconstruct audio from the spectrogram the NTSB launched. It isn’t fairly clear what the board means by saying that it’s going to “look at the scope of the problem and consider options,” however we would not be stunned if it stops importing graphic representations of audio sooner or later.
