NTSB Pulls Thousands Of Crash Dockets After AI Rebuilds Cockpit Audio
The National Transportation Safety Board last week temporarily removed thousands of online investigation dockets after learning artificial intelligence tools could reconstruct cockpit audio from posted spectrogram images.[1]
Spectrogram images from the cockpit voice recorder of UPS Flight 2976, which crashed after takeoff from Louisville in 2025 and killed 15 people, were used to recreate approximate cockpit audio.[1] Federal law bars the agency from releasing cockpit voice recordings, and NTSB spokesman Peter Knudson said the agency is reviewing materials and procedures to protect sensitive onboard audio and video.[1]
The agency learned that AI tools could invert or otherwise reconstruct audio from spectrogram images that the NTSB had posted publicly in investigation dockets.[1] The dockets are used to share technical materials and evidence during probes, but federal law keeps raw cockpit recordings secret.[1]
Privacy and legal experts say the gap between images and recoverable audio is narrowing as AI improves, raising hard questions about how investigators can release useful materials without violating law or privacy.[1]
Show source details & analysis (1 source)
📌 Key Facts
- The NTSB last week temporarily pulled down online public dockets for thousands of investigations after learning AI tools could reconstruct cockpit audio from posted spectrograms.
- Spectrogram images from the cockpit voice recorder of UPS Flight 2976, which crashed shortly after takeoff from Louisville in 2025 killing 15 people, were used to recreate approximate cockpit audio.
- Federal law forbids the NTSB from releasing cockpit voice recordings, and spokesman Peter Knudson said the agency is reviewing materials and procedures to protect sensitive onboard audio and video.
📰 Source Timeline (1)
Follow how coverage of this story developed over time