NTSB Discovers AI-Reconstructed Pilot Voices From UPS Crash Circulating Online
Key Takeaways
- ▸AI voice synthesis tools can reconstruct audio from spectrograms and transcripts, circumventing legal protections designed to keep cockpit voice recordings private
- ▸Data protection regulations often lag behind AI capabilities—format conversions once thought safe are now vulnerable to reconstruction through machine learning
- ▸Government agencies are reassessing how to balance public transparency with privacy and safety in investigations involving advanced AI capabilities
Summary
The National Transportation Safety Board (NTSB) temporarily suspended public access to its accident investigation docket system after discovering that AI voice synthesis tools, including OpenAI's Codex, were used to reconstruct the voices of pilots killed in a UPS plane crash in Louisville, Kentucky last year. The reconstruction was made possible when security researcher Scott Manley highlighted that spectrograms—mathematical visualizations of audio frequencies—published in the public docket could theoretically be converted back into audio. Using these spectrograms combined with publicly available cockpit transcripts, internet users created synthetic approximations of the actual cockpit voice recordings.
The incident exposes a critical gap between existing data protection policies and emerging AI capabilities. While federal law prohibits the NTSB from including raw cockpit audio recordings in its public dockets, the agency had not anticipated that spectrogram images—intended as a safe alternative representation of the data—could be reconstructed into audio by AI voice synthesis tools. In response, the NTSB has restored limited access to its docket system but has closed 42 ongoing investigations for review, including the UPS Flight 2976 case, signaling a broader need to reassess how sensitive information is handled in the era of advanced AI.
