Cops’ favorite AI tool automatically deletes evidence of when AI was used

A recent investigation by the Electronic Frontier Foundation (EFF) has raised serious concerns regarding the use of AI in police reporting. The group claims that Axon's Draft One, an AI-driven tool introduced last summer at a Colorado police department, is nearly impossible to audit, potentially paving the way for dishonest practices in law enforcement. Draft One utilizes a variant of ChatGPT to produce police reports from body camera audio. Officers are expected to review and modify these reports to ensure accuracy and address any biases. However, the EFF argues that the system seems intentionally designed to hinder transparency and accountability. Notably, police departments are not required to disclose the use of AI tools, and Draft One does not save draft versions or document which sections of reports are AI-generated. This lack of record-keeping complicates efforts to compare different iterations of reports, making it challenging for the public to assess the reliability of the technology. The EFF pointedly questions why agencies would avoid maintaining records that could confirm the AI's accuracy. Moreover, the EFF expressed concerns that officers might be hastily approving these reports without thorough review, likening the situation to a 'rubber-stamp' process. This is particularly alarming in light of disclosures from Axon about previous instances where defects in the system allowed officers to bypass safeguards, leading to AI-generated reports being submitted without proper scrutiny. Additionally, there are worries that the AI might misinterpret audio, misrepresenting slang or fabricating context that did not occur.