CrimeRadar App Claims Man Killed In Gillette, But Sheriff Says It Was A Traffic Stop

A man did not die behind a Gillette apartment complex on Wednesday afternoon, despite an automated report on the CrimeRadar app. Campbell County Sheriff's Lt. Dan Maul reviewed the incident on Thursday and confirmed the app's summary was wrong. The report, timestamped 3:18 p.m. MST, categorized the event as a fatality and suggested a possible homicide.

Maul stated that deputies actually conducted a routine traffic stop at that time. They encountered a woman with an outstanding arrest warrant and took her into custody. "No, not even close," Maul said when asked about the app's description. He noted that the audio quality from police radio traffic is often poor, with many codes and abbreviations that can confuse transcription software.

CrimeRadar, which has an estimated 4.6 million Android downloads, uses artificial intelligence to generate summaries of police radio calls. The app includes a disclaimer stating the text is AI-generated and not an official police report. However, the error in Gillette was extreme, transforming a standard encounter into a potential murder investigation. One user commented "WOW" beneath the summary.

Byron Oedekoven, a former Campbell County sheriff and leader of the Wyoming Association of Sheriffs and Chiefs of Police, reacted to the mistake with surprise. "Oh my gosh!" he said. "When an AI glitch can say homicide, I mean that's number one on the big, big crime side." Oedekoven emphasized that AI should be treated as an information tool, not a decision-making one. He advised users to verify details because the technology can confidently invent information.

Researchers have documented similar errors in speech-to-text systems. A 2024 study by Cornell University found that about 1% of transcriptions by OpenAI's Whisper system contained invented phrases. Of those hallucinations, 38% included potentially harmful material such as violent language. The study highlighted how software can supply false information with a straight face, a risk that increases when interpreting noisy police audio.

The Gillette incident is not the first time CrimeRadar has produced false alarms. In August, the app reported a shooting in Sumrall, Mississippi, that did not occur. In Virginia, the app reported an 11-year-old boy overdosing at school in October, though emergency crews found no signs of distress. In Waco, Texas, the app mistakenly identified a training exercise as an officer being shot. CrimeRadar stated on its website that it does not independently verify live dispatch information.

Maul said the sheriff's office did not receive a flood of panicked calls regarding the Gillette report. He encouraged the public to be cautious of AI-generated content without other evidence to verify it. Oedekoven agreed, warning that people will need to double-check information as AI becomes more common. CrimeRadar and the Del Mar Apartments had not responded to inquiries by publication time.