Nana Wu
The integration of Artificial Intelligence (AI) into the machinery of American law enforcement was heralded as a technological panacea. Proponents promised a future where "Draft One" tools—software designed to automatically generate incident reports from body-worn camera (BWC) transcripts—would liberate officers from the drudgery of paperwork, allowing them to return to the streets and focus on community safety.
However, a growing body of independent, peer-reviewed research is challenging the marketing narrative propagated by tech vendors like Axon. Far from being a revolutionary leap forward, current data suggests that AI-assisted reporting is a "solution" in search of a problem. Not only do these tools fail to offer the promised efficiency, but they also introduce significant risks to the integrity of the criminal justice system—risks that may be fundamentally incompatible with the constitutional requirements of due process.
The Mirage of Productivity: A Chronology of Disillusionment
The promise of AI in policing has moved rapidly from hype to implementation, yet the empirical foundation for this transition is remarkably thin.
Early Warning Signs (2024)
The initial excitement surrounding AI-assisted police reports met its first hurdle in 2024. As noted in an ACLU white paper, researchers began investigating whether the software actually achieved its primary goal: saving time. A landmark study published in the Journal of Experimental Criminology found that, contrary to vendor claims, officers using AI tools experienced no significant reduction in the time required to complete their reports.
Expanding the Inquiry (2025–2026)
As adoption increased, so did the scrutiny. In 2025, a report by Walley and Glasspoole-Bird suggested that, rather than saving time, an AI tool in a specific test environment actually increased the time per incident by 18.5 minutes. This trend mirrored findings in other professional sectors, such as software development and clinical medicine, where researchers like Becker et al. (2025) and Rotenstein et al. (2026) found that AI tools often introduced "coordination tax"—the time spent correcting, verifying, and refining AI-generated content often eclipsed the time saved during the initial drafting phase.
The Expert Verdict (The 2026/2027 Study)
The most damning evidence arrived in a study involving 92 senior law enforcement officers—sergeants and above—with an average of 22 years of experience in reviewing police reports. The methodology was straightforward: researchers provided these experts with 80 reports, 20 of which were generated using Axon’s "Draft One" AI.
The results were twofold: first, the experts were unable to distinguish between human-written and AI-written reports at a rate better than a coin flip. Second, when asked to evaluate the reports on clarity, completeness, grammar, accuracy, and utility, the experts found that the AI reports were statistically inferior in the most critical category: accuracy.
Supporting Data: Why "Better" Writing is Not Better Policing
The core of the issue lies in the definition of "quality." AI models are trained to produce fluent, sophisticated, and authoritative-sounding prose. To a layperson, an AI report might look professional. However, the expert reviewers identified a disconnect between stylistic polish and factual precision.
The "Syllable Trap"
Ian T. Adams, a professor of criminology at the University of South Carolina and a former police officer, highlighted that AI often produces writing that is "better" only in the most superficial sense. It increases sentence complexity, uses longer words, and shifts the reading level upward. While this might be ideal for a college essay, it is detrimental to a police report, where the priority is clarity, brevity, and directness.
The Accuracy Deficit
The most alarming finding in the recent study was the statistically significant drop in accuracy scores for AI-assisted reports. According to the research, the use of AI moved a report from the 50th percentile to the 36th percentile in perceived accuracy.
This is not a minor administrative error. In the American criminal justice system, the police report is the foundational document upon which the entire legal edifice is built. It is the primary source material for:
- Prosecutors: Deciding whether to file charges and determining the strength of a case.
- Defense Attorneys: Searching for inconsistencies to protect the rights of the accused.
- Judges and Juries: Reconstructing events months or years after they occurred.
When an AI hallucinates details, misinterprets radio chatter, or smooths over the rough edges of a chaotic encounter, it doesn’t just "improve" the prose—it corrupts the evidentiary record.
Civil Liberties and the "Black Box" Problem
Beyond the failure of utility, there are existential civil liberties concerns that the ACLU has documented in its 2024 white paper. These problems are not merely technical bugs; they are inherent to the way Large Language Models (LLMs) function.
Contamination of Memory
There is a documented risk of "contamination of memory," where an officer, after reviewing an AI-generated draft, may subconsciously adopt the AI’s version of events as their own. This creates a feedback loop where the officer’s testimony in court becomes anchored to a machine’s interpretation of a video rather than their own sensory experience of the event.
The Authoritative Voice
AI models are designed to sound confident. This "authoritative voice" can create a false sense of reliability. Because the reports sound more professional than a typical human-drafted report, they may gain undue influence in the eyes of a jury. If the AI provides more detail—or, worse, adds "flavor" text derived from its training data rather than the specific incident—it risks swaying the judicial process based on algorithmic fabrication.
Official Responses and the Vendor Narrative
Axon and other industry players continue to market these tools under the guise of efficiency and officer well-being. They argue that AI helps reduce burnout by automating administrative tasks.
However, the independent research community is pushing back. As Ian T. Adams noted, there is not a single independent, peer-reviewed evaluation that supports the aggressive marketing claims currently being made by vendors. The industry continues to sell a narrative of "future-proofing" policing, while the reality on the ground—as reported by the very people tasked with auditing these documents—is a decline in the quality of the record.
Implications for the Future of Law Enforcement
The implications of these findings are profound. We are at a crossroads where the pressure to modernize is clashing with the foundational requirements of the legal system.
The "Fun" Factor: Why Officers Might Still Use It
It is worth considering why the technology persists despite its flaws. As one researcher noted, coding a macro or using an LLM to "fix" a data set is often more intellectually stimulating than manual labor. For an officer, watching an AI draft a report can feel like a productive shortcut, even if the time saved is negligible or non-existent. We must distinguish between the experience of using the tool and the actual utility of the tool.
The Call for Human-to-Human Accountability
The consensus among experts is clear: the current generation of AI-assisted police reporting tools is not ready for prime time, and may never be appropriate for the creation of legal documents.
If we value the integrity of the judicial process, we must mandate that police reports remain a human-to-human endeavor. The nuances of a police interaction, the tone of a witness, the chaotic reality of a traffic stop, or the split-second decision-making process during an arrest cannot be captured by a machine that relies on probabilistic text generation.
The drive to deploy AI in contexts where it is clearly inappropriate—driven by marketing power and the allure of "innovation"—threatens to undermine the very system it claims to support. It is time for departments to step back, acknowledge the empirical failures of these tools, and return to the tried-and-true practice of honest, human-led reporting.
The legal system relies on the assumption that an officer’s report is a faithful, independent account of the truth. If we allow that record to be filtered through the opaque, error-prone, and biased lens of an AI, we are not innovating—we are eroding the bedrock of the rule of law.
Siti Muinah
Sagoh
