TL;DR
The US NTSB temporarily shut down its public investigation database after discovering that AI tools enabled individuals to recreate pilots’ voices from cockpit recordings. This development highlights challenges in protecting aviation privacy amid advancing AI capabilities.
The US National Transportation Safety Board (NTSB) has temporarily shut down its public investigation database after discovering that AI and image recognition techniques have been used to reconstruct pilots’ voices from publicly available cockpit data. This move follows the spread of AI-generated audio recreations based on spectrograms from the crash of UPS Flight 2976, which occurred in Louisville, Kentucky, last November.
The NTSB announced on May 21, 2025, that it has suspended access to its online docket system to review how individuals used AI tools to generate audio recordings of cockpit voices from spectrogram images released during the investigation of the UPS Flight 2976 crash. The crash resulted in the deaths of three crew members and 12 on the ground, following a structural failure shortly after takeoff.
The agency clarified that it does not release cockpit voice recordings publicly, citing a federal law enacted in 1990 that restricts such disclosures to protect pilots’ privacy. However, the release of spectrograms—visual sound representations—has enabled some individuals to reconstruct approximate audio using AI models like OpenAI’s Codex and speech processing algorithms. Social media platforms such as X and Reddit have seen examples of these AI-generated recreations.
Experts like Ben Berman, an accident investigator, noted that the combination of advanced AI models and publicly available spectrogram images has made it easier to approximate cockpit sounds, including pilots’ voices, raising concerns about privacy and the potential misuse of such data.
Why It Matters
This development underscores the growing challenge of safeguarding sensitive aviation data in the age of AI. Reconstructed cockpit voices could potentially be misused for malicious purposes or to spread misinformation, complicating efforts to maintain privacy and security in civil aviation. The NTSB’s move to restrict access reflects an attempt to prevent further misuse while it assesses the scope of the issue.
AI voice reconstruction software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background
The NTSB traditionally restricts access to cockpit voice recordings during investigations, with only authorized personnel able to listen to raw data under strict confidentiality. The law passed in 1990 was designed to protect pilots from privacy violations following controversies over public disclosures of cockpit conversations. Recent advances in AI and image recognition, however, have challenged these protections, as spectrograms—images of sound frequencies—are now being used to recreate audio that closely resembles actual cockpit recordings.
The crash of UPS Flight 2976, involving a cargo MD-11F aircraft, is the latest incident where investigation data has been exploited in this manner. The spectrogram of the final 30 seconds of cockpit audio was publicly shared by the NTSB, enabling AI-powered re-creation by individuals within minutes, according to reports.
“All kinds of things are possible now with AI, including reconstructing pilots’ voices from spectrograms, which raises serious privacy concerns.”
— Ben Berman, accident investigator
“The docket system is temporarily unavailable as we examine the scope of the issue and evaluate solutions.”
— NTSB spokesperson
As an affiliate, we earn on qualifying purchases.
What Remains Unclear
It remains unclear how widespread the misuse of spectrogram data is, whether additional investigations have uncovered further instances of AI-generated audio, or what specific measures the NTSB will implement to prevent future misuse. The full extent of potential privacy violations and security risks is still being assessed.
As an affiliate, we earn on qualifying purchases.
What’s Next
The NTSB plans to review its data sharing policies and security protocols before restoring public access to its investigation docket system. Further updates are expected as the agency evaluates technological safeguards and possibly introduces new restrictions on the release of investigation materials.
As an affiliate, we earn on qualifying purchases.
Key Questions
Can AI truly recreate pilots’ voices from spectrograms?
Yes, advances in AI and speech processing algorithms have made it possible to generate approximate audio from spectrogram images, as demonstrated by recent examples on social media.
Why did the NTSB suspend access to its database?
The agency suspended access to review how individuals used publicly released spectrograms and AI tools to reconstruct cockpit voices, aiming to prevent further misuse and protect privacy.
Does this mean cockpit voice recordings are now publicly available?
No, the NTSB does not release cockpit voice recordings publicly. The recent developments involve reconstructed audio based on spectrograms, which are visual representations of sound.
What are the privacy concerns related to this issue?
Reconstructed voices could be used maliciously, spread misinformation, or violate pilots’ privacy rights, raising questions about data security and ethical use of AI in aviation investigations.
Source: Ars Technica
