US scrambles to stop Internet users re-creating dead pilots’ voices

2026-05-22 · Source: AI - Ars Technica · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Emerging Technologies & Innovation · Depth: Fundamental Awareness, medium

Summary

The US National Transportation Safety Board (NTSB) has temporarily suspended public access to its online database of civil transportation accidents after Internet users re-created dead pilots' voices from a spectrogram. This incident involved the November 4, 2025, crash of UPS flight 2976 in Louisville, Kentucky, which killed three pilots and 12 people on the ground. The NTSB had released a spectrogram, a visual representation of sound signals, of the last 30 seconds of cockpit audio during a May 19-20 investigative hearing. Individuals then used software, including the Griffin-Lim algorithm and AI tools like OpenAI's Codex, to reconstruct approximations of the cockpit voice recorder audio. Federal law, enacted in 1990, prohibits the NTSB from publicly sharing cockpit audio to protect air crew privacy. The agency is now reviewing its publicly available materials to prevent similar re-creations.

Key takeaway

For policy makers and legal professionals reviewing data privacy regulations, this incident highlights an urgent need to re-evaluate what constitutes "publicly releasable" information. Your existing policies, designed for human interpretation, may be insufficient against AI-driven reconstruction. Consider updating statutes to explicitly address derived data. You should also assess AI's capacity to infer sensitive information from visual or metadata releases, ensuring privacy protections remain effective.

Key insights

Spectrograms combined with AI tools enable unauthorized audio reconstruction, challenging privacy laws.

Principles

• Federal law protects air crew privacy.
• Visual sound data can reveal audio content.
• AI accelerates data reconstruction capabilities.

Method

Individuals used spectrograms, the Griffin-Lim algorithm, and AI models like OpenAI's Codex to reconstruct cockpit audio recordings from visual sound spectrum imagery.

In practice

• Review public data for unintended information leakage.
• Implement stricter data release protocols.
• Assess AI's role in data re-identification.

Topics

• NTSB Investigations
• Cockpit Voice Recorders
• AI Audio Reconstruction
• Data Privacy
• Spectrogram Analysis
• Federal Aviation Law

Code references

• rbarghou/pygriffinlim
• bkvogel/griffin_lim

Best for: CTO, VP of Engineering/Data, Executive, AI Ethicist, Policy Maker, Legal Professional

Related on AIssential

• AI is being used to resurrect the voices of dead pilots — AI tools can reconstruct sensitive audio from publicly available spectrograms, posing privacy and ethical challenges.
• AI Was Used to Recreate the Voices of Dead Pilots. The NTSB Responded by Locking Down Its Database. — Spectrograms contain reconstructible audio data, enabling AI voice synthesis and prompting NTSB database lockdown.
• ElevenLabs hit with fresh lawsuit over use of voices by Pulitzer and Emmy-winning journalists — AI voice synthesis companies face significant legal risks for unauthorized use of copyrighted voice data.
• Longtime NPR host David Greene sues Google over NotebookLM voice — AI voice replication raises significant intellectual property and likeness concerns for public figures.
• Reconstructing Template-Memorized Images from Natural Prompts — Benign prompts can unintentionally reconstruct template-memorized images, including real people, from generative models with low resources.
• Are AI Systems Incompatible with Data Privacy? — AI systems inevitably infer sensitive personal data from user behavior, creating a fundamental tension with data privacy laws.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI - Ars Technica.