At the recent DEF CON 33 conference, audio engineer Andrew Logan raised concerns about ATC voice communications. “My nightmare is that a sophisticated attacker might use AI voice cloning technology to impersonate air traffic controllers and intermittently give false clearances to cross runways, particularly in adverse weather conditions,” Logan said in his presentation.
Aviation radio is uniquely vulnerable to cloning, Logan explained, because it uses amplitude modulation (AM), is unencrypted, and can be received by anyone with a simple receiver. Handheld aviation transceivers are not expensive, and there have been cases of rogue individuals transmitting false instructions near airports, but that is not the vulnerability that Logan is focusing on. “You need line of sight for that to work,” he explained.
The problem is that powerful air-band transmitters such as the Motorola CM-400 VT VHF are easily available, for example, on eBay for $200. “It’s pretty scary how cheap these things go,” he said. “When paired with an appropriate antenna, it offers an order of magnitude more power, 25 to 45 watts. My fear is that an adversary with that power could transmit much further away from the airfield.”
Logan speculated that the perpetrator could use a hotel room or rental property near an airport to set up the transmitter and operate the equipment remotely from another location or country.
Three parts are involved in the voice cloning process, he said. “First, an encoder finds the unique characteristics of the voice you’re cloning. A synthesizer takes the input text or prompt and applies that voice to it before a vocoder converts that spectrogram that’s generated in part two back into an audible waveform.
“But I don’t believe that an adversary would need to have this level of technical knowledge, use an open-source technology, or roll their own, because there is such a plethora of free or low-cost voice cloning technology available at the moment, and [much] does not have the necessary safeguards.”
Logan shared a voice cloning simulation that he created at JFK Airport in New York City, where a British Airways flight (Speedbird 73A Heavy) was holding short of Runway 22 and communicating on the ground control frequency—on a separate frequency from landing traffic talking to the tower. A JetBlue flight was landing on Runway 22 and normally would land while Speedbird 73 held short. Logan cloned the ground controller’s voice and played it in the video, with the cloned controller saying: “Speedbird 73 Alpha Heavy, cross Runway 22 Right without delay.”
“It only gets better the more you feed it,” Logan said. “And it is trivially easy to create a convincing voice clone, particularly when you’re about to transmit it over a radio that’s going to obscure it further.”
Logan has contacted relevant authorities and entities to warn them about voice cloning, including the FAA, ICAO, the FCC, the U.S. Senate Committee on Transportation, and the National Air Traffic Controllers Association. None has responded to his warning, he said. More recently, however, Logan's presentation was accepted for a session at the Air Traffic Control Association's Technical Symposium in Atlantic City, New Jersey, November 4-6.
“I think that the first thing we can do is get this information in front of as many pilots and controllers as possible. And I hope that’s what I’m doing here today. In conclusion, I really hope that this brings this problem to the attention of pilots, controllers, and regulators.”