90 incidents on record · 2026 Headlights Incident reports by Ellie Harris · Melbourne
10 new this week Library last updated 13 July 2026
← The incident library
HD-INC-088
Consumer AI · United States · 2025 · Unverified automated alerts

An AI crime-alert app turned police-radio chatter into false alarms about serious crimes, and a BBC investigation prompted an apology

By Ellie Harris · Filed BBC Verify investigation reported December 2025

Alleged: CrimeRadar (Scoopz Inc.; linked to NewsBreak) developed or deployed the AI system implicated in this incident. Details are drawn from public reports; parties are presumed innocent of any wrongdoing not established by an official finding.

An AI crime-alert app turned police-radio chatter into false alarms about serious crimes, and a BBC investigation prompted an apology

What happened

It was reported that CrimeRadar, an app published by Scoopz Inc. and linked to the news app NewsBreak, uses AI to monitor publicly available police radio across the United States, transcribe the audio into text, and generate crime alerts that it pushes to residents nearby. In December 2025 a BBC Verify investigation found that the app had sent false alerts about serious crimes to communities in several US states by misinterpreting ordinary radio traffic. In one case in Bend, Oregon, the app’s transcription system misread routine radio traffic as a serious crime and issued an alert that caused alarm. A resident in Minnesota said an alert had defamed him.

It was then reported that CrimeRadar acknowledged what it called serious transcription issues that had led to inaccurate information being sent out, apologised for the distress caused, said it had upgraded its audio processing, and announced plans to let agencies and community members submit corrections and add context to alerts. On the company’s own account, the errors came from how its AI transcribes and summarises radio traffic, and they reached residents as alerts about violent crime in named neighbourhoods before anyone caught them.

What an auditable version would have shown

An alert that a serious crime is happening nearby names a neighbourhood and reaches the people living in it. This one began as a machine transcription of police radio, and the company says that is where the errors came from. An auditable version keeps, for every alert, the original audio, the transcript it produced, how sure it was, and whether anyone confirmed it against a more reliable source before it went out. If all it has is an uncertain transcript, the alert waits. And because the audio and the transcript are kept together, a false alarm can be traced afterwards: a resident who was frightened, or the man who says he was defamed, can be shown exactly what the app heard and what it made of it, and the mistake fixed at the source instead of argued over.

Where the gap was

False alerts about violent crime reached residents, and the company says the errors came from how its AI transcribes radio traffic. A VerificationGate checks a claim like that against a more reliable source, the dispatch record rather than the app’s own transcript, before an alert is sent, so a misheard word does not turn into a neighbourhood scare. A ConductRecord keeps the audio, the transcript and whatever was sent, so a false alert can be traced back to where it started and put right, instead of leaving the people who got it to work out for themselves whether it was ever real.

What governance should have looked like

The bigger the harm an alert can cause, the surer a system should be before it sends one, and an alert about violence is about as big as it gets. A single machine transcription is not a confirmation; it is a lead that still has to be checked. Best practice would be for anything pushing public-safety alerts to confirm each one against a reliable source before it goes out, and to keep the audio, the transcript and the decision on file, so an alert can be checked, traced and corrected instead of taken on the machine’s word.

Failure Pattern: an automated system turned misread audio into public alerts about serious crime, and the errors reached residents before anyone caught them.

Governance Principle: an automated alert to the public should be confirmed against a trusted source before it is sent, not inferred from a single unreliable signal.

The reference implementation of VerificationGate and ConductRecord is open source. It lives at github.com/saffronandindia/headlights-oss, Apache 2.0 licensed and free to install. The repository is public now.

Sources

The mailing list

Fresh incident reports every week. One email to match.

We add new incidents to the library regularly, and send a single short email each week with what's new. The library stays free and open; this is just how you keep up with it.

No tracking. Unsubscribe in one click.

The record

An auditable system would have produced a signed, tamper-evident record the moment this happened: what the system did, the version that did it, the basis it acted on, and the action taken, and CrimeRadar (Scoopz Inc.; linked to NewsBreak) could have produced it on demand.

This is the record the system as deployed did not produce in a signed, auditable form.

What this teaches
Capture what happened when it happens
What the system did, the version that did it, the basis it acted on, and the action taken, recorded at the moment, not reconstructed after.
Sign it, so no one has to trust the record-keeper
A tamper-evident entry. Edit it later and the signature breaks. The record does not ask for the benefit of the doubt.
Make it verifiable by anyone
A court, a regulator, a customer's lawyer can check the record themselves, without taking the company, or us, at our word.

Headlights summarises publicly reported AI incidents. All summaries are independently written, attributed to their original sources, and intended for research and educational purposes. Allegations are identified as such until established through official findings.

Last reviewed June 2026. This report is based on the sources listed above and reflects information available at the time of review; later developments may not be captured. Where a person is described as charged with or alleged to have done something, that allegation is unproven unless a conviction or a court or regulatory finding is stated. Headlights publishes journalism and commentary, not legal advice.

Want to write back?

Direct to my inbox.

ellie@useheadlights.com →