Flock CEO Garrett Langley stood on stage at TechCrunch Disrupt and said his company’s gunshot-detection microphones listen for “just gunshots.” Flock’s own website says otherwise. According to TechCrunch AI, the company’s marketing says the same audio system can flag street takeovers, severe car crashes, fireworks and sideshows. It also says the system’s “detection library is continually expanding.”
This matters well beyond one awkward interview. When an AI audio system is installed on street corners, what it’s trained to recognize is the whole privacy question.
Key Takeaways
- The contradiction is real. Langley said “It’s just gunshots.” A Flock spokesperson later confirmed the system also detects fireworks, sideshows and potential car crashes.
- The company has already walked back one feature. Flock first said it would listen for “screaming.” It then changed that to “sounds of human distress” and later dropped distress detection completely.
- The network is big. Langley said Flock runs about as many microphones as ShotSpotter, which Wired reported at roughly 25,000 in 2024.
- Flock is losing partners. Texas and Florida have moved to end their relationships with Flock, and Amazon’s Ring has ended its partnership.
What Langley Actually Said
Langley framed the product as a public safety tool built for a simple reason. “Most people don’t call 911 when they hear a gunshot, and so that’s why we built the product,” he told TechCrunch editor-in-chief Connie Loizos.
He said the idea of going beyond gunfire came from customers. Some had microphones in parking lots and wanted alerts “if someone was being abducted, if someone was being harmed.” Flock built a prototype, tested it with a few customers, and ran into a wave of concern. Langley admitted the company made mistakes and said it settled on a simple rule: “it’s just going to do gunshots, plain and simple.”
He also pushed back on what he called misinformation, arguing that critics don’t understand how the products work. Protesters gathered outside the San Francisco venue while he spoke.
The Company’s Clarification
Flock spokesperson Paris Lewbel gave TechCrunch a more precise version. The technology is “primarily used to detect gunfire but is also capable of detecting fireworks, sideshows, and potential car crashes.” Lewbel said Langley meant to stress that the system “is not designed to monitor conversations or analyze speech and does not currently detect human screaming or distress.”
That’s a narrower claim than “just gunshots,” and the word “currently” is doing a lot of work in it.
Why This Matters for AI
What stands out here is how easy it is to quietly expand the scope of an AI classifier. Once the microphones are on the poles, adding a new sound category is mostly a model update. No new hardware, no new installation, and often no public vote.
That’s the core tension in AI-powered surveillance:
- Scope creep is cheap. A gunshot detector and a “disruptive events” detector can run on the same hardware. The difference lives in the software.
- Marketing and assurances drift apart. Sales materials push capability. Public statements push restraint. Communities end up unsure which one is true.
- Customers drive the roadmap. Langley said customer requests led to the distress-detection prototype. That’s normal for SaaS, but these customers are mostly law enforcement agencies, and the people being recorded never get a say.
This isn’t new territory. ShotSpotter has faced years of questions about accuracy and oversight, and some cities have cut ties with it. Flock’s license plate readers already draw criticism from the ACLU and other privacy advocates, and some people have gone as far as destroying its cameras.
What to Watch
- Procurement scrutiny. Expect city councils and state agencies to ask for written limits on which sounds a system can detect, not verbal promises.
- Contract churn. Texas, Florida and Ring pulling back suggests more customers could follow if trust keeps slipping.
- Regulatory pressure. Clear disclosure rules for AI audio detection in public spaces look more and more likely.
For anyone building or buying AI sensing tools, the lesson is plain. Your public claims, your marketing copy and your model’s actual capabilities need to match. When they don’t, someone will notice, and in this case they noticed live on stage.
The full interview and Flock’s statement are in the original TechCrunch AI report.