Flock CEO says its microphones detect ‘just gunshots’ — its website says different

The Trust Gap in Public Safety Tech

As a builder, you learn early on that your word is your most valuable asset. If you tell a customer your software does one thing, but the code—or the marketing team—says another, you aren't just dealing with a bug. You're dealing with a breach of contract with the public. This is the situation Flock Safety currently finds itself in, and it serves as a masterclass in how not to handle the rollout of sensitive AI hardware.

Flock has built a massive business on automated license plate readers, but they’ve recently expanded into audio detection. The pitch is simple: sensors that hear a gunshot and alert police instantly. In a recent public defense of the technology, CEO Garrett Langley was adamant that these microphones are tuned for one thing and one thing only: the sound of gunfire. He claimed the devices aren't recording ambient conversations or listening for other noises. The problem is that his own company’s website and documentation have, at various times, suggested otherwise.

Marketing vs. Reality

When you look at the technical collateral Flock has put out, the scope seems much wider than "just gunshots." Documents have pointed to the detection of glass breaking, tires screeching, and even human screaming. For a privacy advocate, that’s a red flag. For a founder, it’s a messaging nightmare. You cannot claim a device is a single-purpose tool while simultaneously advertising its multi-purpose capabilities to procurement officers.

This discrepancy matters because of the environment we’re building in. We are currently in a transition period where the public is deciding how much they trust AI-powered surveillance. When a leader says "it only does X," but the fine print says "it can also do Y and Z," the public assumes the worst. They assume Y and Z are the real goals, and X is just the foot in the door.

The Surveillance Scope Creep

Every builder knows about feature creep. You start with a simple MVP, and before you know it, you’ve added ten features no one asked for. In the world of government tech and surveillance, feature creep turns into scope creep, which quickly turns into unconstitutional monitoring. If a microphone is sensitive enough to pick up a tire screech, it is sensitive enough to pick up a loud conversation on a sidewalk.

Flock’s defense is likely that the AI filter sits on the edge, only triggering an alert when a specific waveform is identified. But the history of tech tells us that once the hardware is in the ground, the software can be updated. A "gunshot only" sensor today is a "protest monitor" tomorrow with a simple firmware push. This is why the ACLU and other privacy groups are breathing down their necks. They aren't just worried about what the tech does today; they are worried about the lack of transparency regarding its actual ceiling.

The Founder’s Perspective: Radical Honesty

If I’m building in this space, I have to be more than just compliant. I have to be radically honest. If my device can hear screaming, I need to say it hears screaming and explain why that’s a net positive for the community—and, more importantly, what I’m doing to ensure that data isn't being abused. Trying to downplay capabilities to avoid heat always backfires. It makes you look like you’re hiding something, even if you’re just trying to simplify a complex pitch.

The tech industry has a bad habit of using "security" as a blanket excuse to bypass transparency. We’ve seen this with AI training sets and we’re seeing it now with IoT hardware in our cities. Langley’s insistence that the tech is limited doesn't hold water when the sales deck says the opposite. It suggests a disconnect between the executive suite and the product team, or worse, a deliberate attempt to pacify the public while selling a more intrusive product to the state.

• Precision is everything: If you say your AI only detects one thing, ensure the model is literally incapable of detecting others.
• Alignment matters: Your marketing site, your CEO's mouth, and your technical documentation must say the same thing.
• Privacy by design: In public safety, if it can be abused, the public will assume it is being abused. You have to build in the guardrails from day one.

What This Means for the Future of Smart Cities

We are at a crossroads. We can have safer cities through technology, or we can have digital panopticons. The difference lies in the builders. If we continue to see companies like Flock trip over their own narratives, the pushback from the public will eventually lead to blanket bans on the very tech that could actually save lives. Over-promising to the police while under-sharing with the public is a recipe for a regulatory crackdown.

The most dangerous thing a founder can do in the AI era is treat the public like they won't read the fine print. They will. And they'll bring receipts.

The lesson here is simple: Don't build a "black box" and expect people to trust the color you painted it. If the microphones are listening for more than shots, say so. Explain the trade-offs. Otherwise, you’re just building a target for the next class-action lawsuit or city council ban. In the world of AI surveillance, your reputation is the only thing that doesn't get a firmware update.

Read the original at TechCrunch AI →