NTSB Halts Online Docket After AI Recreates Pilot Voices

The National Transportation Safety Board announced on May 21 that its online docket system is temporarily unavailable. The move follows a wave of internet users who used image‑recognition software and AI to synthesize pilots’ last words from publicly released sound‑spectrum images.

The board’s statement warned that advances in computational methods now let individuals approximate cockpit voice recorder (CVR) audio from the visual data the NTSB already publishes. The UPS Flight 2976 crash investigation in Louisville, Kentucky, provided the most recent example. The board reiterated that it never releases raw CVR recordings.

A 1990 federal statute explicitly prohibits the NTSB from sharing any cockpit voice or video recordings. The law was enacted after pilots objected to a televised broadcast of cockpit dialogue from a 1988 Delta Air Lines crash. Its purpose is to protect aircrew privacy and prevent misuse of sensitive audio.

In the UPS incident, an MD‑11F cargo plane lost an engine moments after takeoff on November 4, 2025, killing three pilots and twelve people on the ground, while injuring 23 others. The NTSB’s investigation released spectrograms—visual representations of sound frequencies—without the underlying audio, assuming they could not be reverse‑engineered.

However, AI models trained on public audio datasets can now infer speech patterns from those spectrograms, producing convincing voice recreations. The resulting clips spread across social media, raising concerns about privacy, misinformation, and the integrity of official investigations.

By suspending public access, the NTSB aims to halt further reconstructions while it reviews how to present technical data without enabling audio synthesis. The board’s action underscores the tension between transparency in safety investigations and the unintended consequences of emerging technology.

What to watch next: The NTSB is expected to issue new guidelines on publishing spectrographic data and may propose legislative updates to address AI‑driven reconstruction risks.