Flock drops 'screaming detection' pilot under EFF pressure, but its audio fleet stays deployed

Flock Safety has scrapped a limited audio-detection pilot that listened for sounds of "screaming" in public spaces, the Electronic Frontier Foundation announced Friday. The cancellation is a narrow win for an unusual coalition of camera saboteurs, EFF advocates, and online organizers.

Flock, a license-plate-camera company that has been expanding into broader surveillance products, described the audio-detection system on its company blog as a "human distress detection" capability that ran as an always-on microphone on certain devices, classifying sounds as "screaming" and forwarding matches to customers, often local police. The company characterized the pilot as a "limited trial" that was "never broadly released" and ended "after careful consideration and community consultation."

That framing is partly accurate. The pilot was narrow. It is also partly a pre-built exit ramp. Flock has thousands of audio devices deployed across U.S. communities, frequently at the request of local law enforcement and often without public notice. The screaming-detection feature is gone; the underlying audio infrastructure is not.

The specific mechanism that made this feature uniquely killable: always-on recording of human voices in public sits in a different legal and reputational category than license-plate recognition, which captures only strings of characters on vehicles already subject to public registration. Audio of voices implicates First Amendment assembly and association in a way plates do not, and it produces the kind of visceral backlash that "limited trial" framings cannot easily absorb. The corporate-friendly "small pilot" language that works for, say, a new analytics dashboard does not work as well when the public-facing description is "we are listening for screaming." Flock's pre-built exit lane, "this was always a small thing we tried," was wide enough to drive the feature through, and only that feature.

The cancellation followed months of multi-vector pressure. Anonymous actors cut down, blinded, and disabled Flock cameras; EFF and allied groups have defended them as "folk heroes," with legal-defense funds established to support them. Organized online opposition built a sustained negative-information environment around the company's audio expansion. EFF applied direct institutional pressure, framing the pilot as an unprecedented civil-liberties escalation: an always-on microphone in public space that could give law enforcement pretext to swarm any community on the basis of an algorithm's guess.

The asymmetric-cost logic predicts the partial outcome. Killing a single experimental feature preserves the broader audio business, which is where the actual revenue and the actual law-enforcement relationships live. Announcing the broader audio-device fleet as a casualty would have meant conceding the entire expansion. The narrow concession is the rational one for the company, and the genuine win for the opposition.

EFF called the cancellation a "Victory" and used the moment to argue that "public pressure can sometimes work to influence" corporate surveillance decisions. That framing is honest as far as it goes. The strongest counter-reading is that Flock was retiring the feature for business reasons, including low customer uptake, technical issues, or competitive priorities, and EFF is retrofitting causation onto a pre-existing decision. TechTimes reported the cancellation under the headline "Cameras Still Listen," which captures the partial-win reality.

The watch item is what happens to the rest of the audio fleet. If Flock treats the screaming-detection cancellation as a settled matter and continues to expand audio-device deployment under existing local-police contracts, the narrow win is just a feature-level footnote. If the cancellation opens space for renewed pressure on the always-on audio architecture itself, the same mechanism that killed one feature could be turned against the broader deployment, but only if the coalition that found "screaming detection" uniquely indefensible can extend the argument to the microphones that stayed.