NTSB Limits Docket Access After AI Reconstructs Pilots' Voices
AI tools have been used to recreate voices of deceased pilots from cockpit voice spectrograms, prompting NTSB to restrict access.
Why it matters: The use of AI to resurrect voices from cockpit spectrograms highlights privacy gaps and legal risks with posthumous data use. For privacy and litigation counsel, the incident exposes new liabilities around digital evidence and deepfake potential.
- AI models recreated deceased pilots' voices using only CVR spectrogram images, bypassing audio files.
- NTSB confirmed a temporary ban on public access to some docket systems as of May 20, 2026.
- Concerns focus on privacy rights of victims and risk of AI-generated voices being mistaken for original evidence.
- Experts say current privacy laws may not cover this type of posthumous voice data use.
AI technology has advanced to the point where a deceased pilot’s voice can be reconstructed using nothing but publicly available cockpit voice recorder (CVR) spectrograms. The process, which requires no audio files, has recently allowed researchers to create close replicas of original pilot voices from image files alone.
- The National Transportation Safety Board (NTSB) responded by temporarily restricting access to docket systems containing CVR spectrograms, aiming to
safeguard parties and victims
, according to NTSB Public Affairs Officer Eric Weiss. - NTSB dockets generally exclude original audio due to privacy rules, releasing only transcripts and sometimes spectrogram images. However, thousands of spectrograms spanning major investigations remain in the system.
- The capability to restore a person’s voice from a simple image file “raises new questions about the adequacy of current privacy laws,” noted law professor Orin Kerr.
- NTSB Chair Jennifer Homendy emphasized the challenge of maintaining transparency for safety investigations without compromising the privacy of families and deceased pilots.
- Concerns have been raised about AI-generated voices being confused with authentic evidence in investigations or legal disputes.
While the technical specifics of these AI models have not been fully disclosed, the demonstration of voice recreation with only a single image underlines the need for debate around digital privacy, posthumous rights, and deepfake risks. There is no public record yet of litigation or regulatory investigation sparked specifically by these reconstructions.
Read more about the legal and privacy implications in major coverage from TechCrunch and AIN Online.
By the numbers:
- Thousands — CVR spectrogram images in NTSB dockets.
- 1 — Number of image files needed to recreate a pilot’s voice with AI.
- May 20, 2026 — Date NTSB imposed temporary docket access restrictions.
Yes, but: The fidelity of voice reconstruction via AI remains difficult to verify, as full technical details are not public.
What's next: NTSB is revisiting its disclosure policies for CVR data and plans to issue updated guidance.