Flock’s CEO says its microphones detect ‘just gunshots’, but its website also lists crashes and sideshows

What did Flock’s CEO say?

Flock Safety, the Atlanta-based company best known for automated licence-plate readers, also sells acoustic sensors that listen for gunfire. Interviewed by TechCrunch editor-in-chief Connie Loizos at Disrupt 2026 in San Francisco, chief executive Garrett Langley said those microphones detect “just gunshots”, and pushed back on what he called misinformation about the product.

“Most people don’t call 911 when they hear a gunshot, and so that’s why we built the product,” Langley said. He explained that some customers had asked for more: “Oh, we have these in parking lots. It would be great if we could get alerted if someone was being abducted, if someone was being harmed.” Flock prototyped that with a couple of customers, he said, then pulled back after concerns were raised. “OK, great, it’s just going to do gunshots, plain and simple. If a gun goes off, we’ll notify, and if not, that’s all it’s going to listen to.”

Asked about the gap between that answer and the company’s marketing, he repeated: “It’s just gunshots.” Protesters gathered outside the venue during the interview, TechCrunch reported.

What Flock’s own pages say the sensors detect

Flock’s Audio Detection product page describes a broader system. It says the sensors “detect gunfire and critical sounds,” promises to “detect street takeovers early to stop chaos before it escalates” and to “identify severe crashes in real time so help arrives faster,” and offers to “expand awareness beyond gunfire with alerts for car crashes, fireworks, and sideshows.” A sideshow is an illegal street gathering where drivers perform stunts.

After the interview, Flock spokesperson Paris Lewbel gave TechCrunch a statement that matches the website rather than the stage answer: “Flock’s gunshot detection technology is primarily used to detect gunfire but is also capable of detecting fireworks, sideshows, and potential car crashes. CEO Garrett Langley’s remarks were intended to emphasize that Flock’s Audio Detection technology is not designed to monitor conversations or analyze speech and does not currently detect human screaming or distress.”

TechCrunch also quoted an FAQ on Flock’s site saying the detectors can pick up “fireworks, sideshows, and other disruptive community events,” and that the “detection library is continually expanding to include emerging threats.” So the dispute is less about whether the sensors hear voices than about what “just gunshots” means when the product is sold as a growing library of sound events.

How Flock says the microphones work

Flock set out its account of the technology in a blog post by chief product officer Baily Quintrell, first published in July and updated in August. It says the detectors analyse sound continuously in rolling windows of about three seconds, using a low-power processor on the device that asks one question: does this sound match a qualifying event, “such as a gunshot, vehicle crash, or tires screeching?”

If not, Flock says, the audio is never recorded, stored or sent anywhere. If so, the device records, encrypts and saves only a short clip, estimates the location using detections from several sensors, and sends an alert. The post says no customer or Flock employee can open a live audio feed, that the system cannot understand speech or identify voices, and that a clip may contain a few incidental words spoken near the triggering sound. Limited, pre-scheduled diagnostic samples are also retained to improve the system.

Like other on-device classifiers, such as the Gemma scam detector Google DeepMind built with Singapore, the privacy case rests on processing locally and discarding what does not match. Flock’s version is tied to its camera network: a confirmed event can surface vehicles seen entering or leaving the area on nearby licence-plate readers. Flock says that is a starting point for investigators, not an automatic enforcement action, and that clips from wrong alerts are fed back to improve the model.

The screaming feature Flock removed

The sensitivity over voices dates to 2025. When Flock began promoting a “distress” alert for what was then called Raven, the Electronic Frontier Foundation noted that an online ad showed police receiving an alert for “screaming,” and that Flock later changed the wording to “distress.” The EFF warned that the feature could run into state eavesdropping laws and urged cities to cancel Flock contracts.

Flock’s blog says the human-distress feature, including screaming, was a pilot available to a small number of customers and was never broadly released, and that after “careful consideration and community consultation” the company removed it. Langley’s account on stage was similar: a prototype with a couple of customers that was dropped. The spokesperson’s phrase, that the system “does not currently detect” screaming or distress, leaves the door open in a way the CEO’s answer did not.

Why the wording matters

Acoustic gunshot detection already has a contested record. Critics including the EFF point out that such systems often mistake car backfires or fireworks for gunfire. Flock’s own materials acknowledge that “no acoustic system is perfect” and present fireworks detection partly as a way to cut false gunshot alerts.

Each new sound category changes what the sensors are for, however. Crash and sideshow alerts turn a gunfire tool into a general street monitor whose outputs can be joined with licence-plate records. That is why the precise list matters to the cities that buy the devices and to residents living under them, much as buyers of home devices with always-ready far-field microphones want to know exactly when they listen.

Langley said Flock’s microphone count is comparable to ShotSpotter’s, which Wired put at about 25,000 in 2024. Flock did not give its own figure.

A company under pressure

The interview came during a difficult stretch for Flock. In a separate part of the Disrupt conversation, Langley said the company has not ended a contract with any police department over misuse of its licence-plate data, despite reports of officers disciplined for abusing access, though he said that is not off the table. TechCrunch puts Flock’s network at more than 100,000 licence-plate cameras in the US.

Days earlier, TechCrunch reported, citing Reuters, that Flock was reducing its workforce by about 18%, or roughly 270 people, through a voluntary buyout. TechCrunch also reported that Texas and Florida have moved to end their relationships with the company and that Amazon’s Ring ended its partnership earlier this year.

What remains unclear

The Flock pages AI Lookout read do not give a full list of the sound events its sensors classify, how often each is triggered, or independent accuracy figures for its audio models. Nor has it said what process would be followed before a new category, such as distress, is added again. Device-level protections like Google’s AISeal enclave for on-device AI on Android show how vendors can document what local models can and cannot access; Flock’s guarantees so far rest on its own descriptions.

For cities weighing a contract, the practical step is to read the product terms rather than the stage answer: ask which event types are switched on, how long clips are kept, and who can link an audio alert to plate data. By Flock’s own account, the answer is more than gunshots.