Spectre I aims to jam AI wearables that now defeat noise jammers

The Atlantic traces an escalating arms race between AI-enabled recording wearables and the countermeasures meant to defeat them. The piece opens with Anthony "Bingy" Arillotta, who in August 2003 was strip-searched and handed a bathrobe before a Genovese crime family induction meeting, a level of paranoia the article argues ordinary people may soon need too, as sensors built into glasses, necklaces and lapel pins can silently record conversations. Apple, described as having the largest personal-tech ecosystem in the world, is rumored to be developing an AI pin or pendant that would act as an iPhone's constant eyes and ears, and other companies are working on similar wearables.

In March, the start-up Deveillance announced Spectre I, a hockey-puck-shaped device meant to stop others from recording nearby conversations. Deveillance was founded by Aida Baradari, described as a recent college graduate who grew worried about the rise of AI-enabled recorders. Baradari would not explain exactly how Spectre I's jamming signal works, saying only that it resembles speech. The product's launch video also claims Spectre I will detect nearby microphones, but when asked how, Baradari said her team is still "working on that part right now."

Older jammers relied on loud white noise or, more recently, ultrasonic tones that exploit a quirk in microphone hardware. In 2020, a University of Chicago team led by Yuxin Chen mounted 23 ultrasonic transducers on a single bracelet so jamming signals could radiate in every direction instead of toward one target. But the article argues these approaches are losing ground: modern AI wearables use speech-recovery algorithms, the same class of technology behind hearing aids and noise-canceling headphones, to strip out both ambient noise and ultrasonic jamming and reconstruct speech, even inferring missing syllables from context. DeLiang Wang, a computer scientist at Ohio State University, has spent decades training neural networks on hundreds of hours of recorded voices to build this kind of noise suppression for hearing aids, and Microsoft has run an annual Deep Noise Suppression Challenge since 2020 to advance the field for products like Teams.

The article surveys other obfuscation-based countermeasures: Berlin-based artist Adam Harvey's makeup and clothing designed to confuse facial-recognition systems; the browser extension TrackMeNot, built by Daniel Howe and Helen Nissenbaum, which floods a user's Google search history with randomized decoy queries; and "babble tapes", audio layered with 40 tracks of voices in different accents, which defense attorneys once played during jailhouse client meetings, a technique described to the author by Boston University law professor Woodrow Hartzog. In 2023, a team led by Ming Gao, now at Nanjing University, built a wearable jammer called MicFrozen that listens to its wearer and generates real-time ultrasonic "anti-speech" tuned to that person's voice, then layers on counterfeit speech-shaped sound meant to mislead any algorithm trying to reconstruct what was jammed.

The article closes on two notes. First, speculation that future AI wearables could sidestep audio jamming entirely, reading lips through a window the way HAL 9000 does in 2001: A Space Odyssey, or in principle recovering speech from vibrations on the surface of a glass of water. Second, a return to the Mafia's own surveillance arms race: after crime families cycled through bugged phones, safe houses, burner phones and encryption, the FBI in 2018 began secretly running its own encrypted-phone company, Anom, selling 12,000 devices through informants to criminal organizations worldwide while intercepting every message sent through them. Hartzog frames the wider AI wearable dynamic in similar terms, warning that the countermeasure side of the fight is vastly outmatched.

Key facts

Deveillance, founded by Aida Baradari, announced the anti-recording device Spectre I in March; Baradari declined to explain how its jamming signal works beyond saying it resembles speech, and said microphone-detection, a feature claimed in the launch video, is still being developed.
AI wearables now use speech-recovery algorithms, the technology behind hearing aids and noise-canceling headphones, to defeat both ambient noise and ultrasonic jammers by reconstructing speech and inferring missing syllables from context.
In 2020, a University of Chicago team led by Yuxin Chen built a jamming bracelet with 23 ultrasonic transducers; in 2023, a team led by Ming Gao built MicFrozen, a wearable that generates real-time ultrasonic anti-speech tuned to its wearer's voice.
Apple is rumored to be developing an AI pin or pendant serving as an iPhone's constant eyes and ears, part of a wider wave of AI-enabled recording wearables the article says could become as common as AirPods.
The FBI ran its own encrypted-phone company, Anom, from 2018, secretly selling 12,000 devices to criminal organizations through informants while intercepting every message sent.

Why it matters

Recording paranoia used to be the province of spies and mobsters, like Anthony Arillotta getting strip-searched before a 2003 Mafia induction meeting. The article's argument is that AI-enabled wearables, plus a rumored Apple AI pin or pendant, are pushing ambient recording toward the mainstream, and that the speech-recovery algorithms behind these devices have gotten good enough to defeat the jamming techniques ordinary people have relied on for years.

Who it affects

Anyone having a private conversation near someone wearing AI-enabled glasses, a necklace or a lapel pin, since these devices can serve as silent notetakers, assistants or informal therapists. It also affects the small pool of researchers and start-ups, among them Deveillance, working on countermeasures against a technology backed by the much larger speech-processing industry that also serves hearing aids, smart speakers and teleconferencing.

How to use it

Spectre I is a hockey-puck-shaped device from Deveillance meant to stop nearby devices from recording usable audio. Founder Aida Baradari says its jamming signal resembles speech, and its launch video claims it will also detect nearby microphones, though Baradari told the author that capability is still being developed. The source gives no price or ship date for Spectre I.

How solid is it

The reporting draws on named academic sources: DeLiang Wang at Ohio State on neural-network speech recovery, the University of Chicago's 2020 ultrasonic-bracelet jammer led by Yuxin Chen, and Nanjing University researcher Ming Gao's 2023 MicFrozen jammer. Spectre I itself is thinner: Baradari would not describe the actual jamming mechanism and confirmed the microphone-detection feature is unfinished, so its real-world effectiveness is untested in the source.

Risks and caveats

The source does not say whether Spectre I has shipped, what it costs, or how its jamming signal actually works technically, only that it resembles speech. It is also silent on whether the University of Chicago bracelet or MicFrozen were ever commercialized outside research settings, and on Apple's AI pin or pendant, which remains an unconfirmed rumor. Hartzog's closing point is structural: countermeasure researchers are vastly outfunded by the corporations building the recording technology they are trying to defeat.

“The thing about cat-and-mouse games is that we know how they usually end up for the mouse. And in this case, the cat includes some of the most powerful corporations to ever exist.”

— Woodrow Hartzog, law professor at Boston University