The Milwaukee Police Department is considering how generative artificial intelligence could be used in its work, but the department says it is not currently using the technology and is nowhere close to deploying it. MPD adopted a generative AI policy in January. The rules say AI-produced material must be treated as a draft, checked for accuracy and labeled in the final work product when the technology has been used. The department’s Generative Artificial Intelligence Committee has reviewed potential products while assessing their usefulness, cost, accountability and possible effects on police work, an MPD spokesperson said. One possible application would be helping officers prepare reports from body-camera audio. Axon, the company that supplies MPD’s body cameras, markets a product called Draft One that uses audio to create draft police reports. The technology’s potential use in police documentation has raised concerns among civil rights advocates and defense attorneys, who say errors, omissions and bias could affect people accused of crimes. “There is this risk of errors and hallucinations and bias that we see a lot of large language models have,” said Amanda Merkwae, policy and advocacy director for the American Civil Liberties Union of Wisconsin. She said problems could include transcription mistakes or bias embedded in the data used to train AI systems. The Milwaukee County District Attorney’s Office has also considered how AI could affect criminal prosecutions. Jeffrey Altenburg, the office’s chief deputy district attorney, said the technology might be useful if people maintain meaningful oversight and verify that its output matches what officers intended to report. But AI-generated text could not replace an officer’s direct observations, said Deputy District Attorney Matthew Torbenson. In a child-neglect investigation, for example, details such as a home’s appearance and smell, insect infestation or whether food is available would have to come from officers at the scene. The consequences of an inaccurate report could extend beyond a single charging decision. Merkwae questioned what would happen if an officer’s later testimony conflicted with a report produced or assisted by AI. Such contradictions could raise questions about whether an officer’s credibility problems must be disclosed to defendants through a prosecutorial Brady list, which tracks law enforcement witnesses whose credibility issues may be relevant to a criminal case. Connor Romenesko, a forensic science practice coordinator with the Wisconsin State Public Defender, said problems could occur even without intentional misconduct. An officer finishing a shift might fail to thoroughly review an AI-generated draft, and prosecutors could then rely on inaccurate information when deciding whether to charge someone. MPD’s policy signals that the department is preparing for possible future use while requiring human review. For now, however, officials say any technology adoption remains under consideration rather than an imminent change in how Milwaukee police document their work.