AI is being used to resurrect the voices of dead pilots

TL;DR

AI has been used to recreate the voices of pilots who died in a 2025 UPS plane crash. The NTSB responded by temporarily removing related investigation files from public access. The development highlights emerging challenges in handling sensitive data with AI.

The National Transportation Safety Board (NTSB) has temporarily restricted public access to its investigation docket after discovering that voices of pilots killed in a 2025 UPS plane crash were recreated using artificial intelligence and circulated online.

The incident involves the investigation of UPS Flight 2976, which crashed in Louisville, Kentucky, in 2025, resulting in the deaths of the pilots. This case highlights the broader issues of AI-generated content in aviation safety. The NTSB confirmed that a spectrogram file from the cockpit voice recorder, which is normally protected from public release by federal law, was publicly available in the investigation docket.

Using publicly available data—including the spectrogram image and a transcript—the AI community, including users on social media, reconstructed audio approximations of the pilots’ voices. Reports indicate that AI tools like Codex were employed to generate these voice recreations, although the exact methods and extent of the AI’s involvement remain under review. The NTSB has acknowledged that this use of AI to recreate voices raises significant safety and ethical concerns, prompting the agency to temporarily restrict access to the investigation docket while it reviews the situation.

Why It Matters

This incident underscores the growing capabilities of AI to reconstruct sensitive audio, even from limited data, raising questions about privacy, safety, and the integrity of accident investigations. For the aviation industry, it highlights the need for stricter controls over cockpit data and the potential for AI to be misused to produce misleading or damaging content. For the public and regulators, it emphasizes the importance of establishing clear boundaries and safeguards around AI-generated reconstructions of critical safety data.

AI Voice Recorder with Playback, 5'' Large Touch Screen Digital Recorder, 13 Languages Transcription, 8 Mics Noise Reduction, 2600mAh Battery, AI Summary & Translation for Meetings, Lectures

AI Voice Recorder with Playback, 5'' Large Touch Screen Digital Recorder, 13 Languages Transcription, 8 Mics Noise Reduction, 2600mAh Battery, AI Summary & Translation for Meetings, Lectures

【Smart Voice Recorder Transcriber】This voice recorder with transcription to text uses cutting-edge AI, offering unlimited free, fast and…

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Background

The NTSB has long maintained that cockpit voice recordings are classified and protected, with access limited by law. This underscores the importance of safeguarding sensitive investigation data. The 2025 UPS crash investigation was among the most recent in which such data was included in the public docket, which is typically open to transparency. The spectrogram file, a visual representation of sound signals, was publicly accessible and used as the basis for AI-generated audio. This development follows broader concerns about AI’s ability to recreate realistic audio and video content, which has become a topic of debate in various sectors, including aviation safety and digital security.

“We are reviewing the circumstances that led to the reproduction of pilot voices using AI and are assessing the security protocols for sensitive investigation data.”

— NTSB spokesperson

“It’s technically feasible to reconstruct audio from spectrogram images using AI, but applying this to sensitive safety data raises serious ethical questions.”

— Scott Manley, YouTuber

OcenAudio Software User Guide for Beginners: A Complete Step-by-Step Guide to Editing, Enhancing, Mixing, and Mastering Audio; From Absolute Beginner to Skilled Producer

OcenAudio Software User Guide for Beginners: A Complete Step-by-Step Guide to Editing, Enhancing, Mixing, and Mastering Audio; From Absolute Beginner to Skilled Producer

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

What Remains Unclear

It remains unclear how widespread the circulation of these AI-generated voices is, whether any official investigation or legal action is planned, and what specific safeguards will be implemented moving forward. The full extent of the AI’s capabilities in this context is still being evaluated by the NTSB and cybersecurity experts. Understanding these capabilities is crucial for future safety protocols.

AI Voice Recorder, Voice Recorder w/Case Display Haptics, Transcribe & Summarize&Translate with AI Technology, 118 Languages Support, 64GB, Audio Recorder for Lectures, Meetings, Calls, Black

AI Voice Recorder, Voice Recorder w/Case Display Haptics, Transcribe & Summarize&Translate with AI Technology, 118 Languages Support, 64GB, Audio Recorder for Lectures, Meetings, Calls, Black

AI Voice Recorder, Voice Recorder w/Case Display Haptics, Transcribe & Summarize&Translate with AI Technology, 118 Languages Support, 64GB…

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

What’s Next

The NTSB is expected to complete its review of the incident within the coming weeks, which may include tightening access controls to sensitive data and issuing new guidelines on AI use in accident investigations. Regulations around AI in safety investigations are likely to evolve. Further technical assessments are also anticipated to determine the scope of AI’s ability to recreate audio from visual data like spectrograms.

128GB Magnetic Voice Activated Recorder - Digital Voice Recorder with DSP 5.0 Triple AI-Intelligent Noise Cancellation, 9800Hrs Dictaphone with Playback for Lecture Meeting, Recording Device, MP3

128GB Magnetic Voice Activated Recorder – Digital Voice Recorder with DSP 5.0 Triple AI-Intelligent Noise Cancellation, 9800Hrs Dictaphone with Playback for Lecture Meeting, Recording Device, MP3

【Magnetic Voice Recorder】Featuring a strong built-in magnet, this voice recorder with playback capability can be effortlessly attached to…

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Key Questions

How did AI recreate the voices of the pilots?

AI tools, such as Codex, used the publicly available spectrogram images and transcripts to generate approximate audio representations of the pilots’ voices.

Why did the NTSB temporarily remove access to the investigation docket?

The agency took this step to review security concerns after discovering that sensitive cockpit voice data could be reconstructed and circulated using AI, potentially compromising investigation integrity.

Are these AI-generated voices accurate representations of the pilots?

It is not yet confirmed how precise or authentic the reconstructed voices are; experts say they are approximations based on limited data.

Could this happen in other investigations?

Yes, the incident raises concerns that similar AI reconstructions could occur in other cases involving limited or visual data, prompting calls for stricter data protections.

What are the ethical concerns surrounding this development?

The main issues include privacy violations, potential misinformation, and the impact on the integrity of official investigations.

Source: TechCrunch

You May Also Like

Glasspane: When Transparency Itself Becomes the Product

Glasspane has detailed new workforce, AI telemetry and public sharing features for infrastructure visibility.

DuckDuckGo search saw 28% more visits after Google said people love AI mode

DuckDuckGo experienced a 28% increase in search visits following Google’s claim that users love AI mode, highlighting user preference for privacy-focused search.

sp.h: Fixing C by giving it a high quality, ultra portable standard library

The new sp.h library offers a minimal, portable, high-quality C standard library built directly against syscalls, bypassing libc dependencies.

The Defender’s Window Is Closing Faster Than Anyone Is Counting

April 2026 reports point to faster AI-assisted security fixes, attack automation and open-weight model gains.