AI Generated Police Reports | What We Need to Know

Can Artificial Intelligence Write Police Reports? What We Need to Know

Artificial intelligence (AI) is increasingly being introduced into policing and the legal system across North America. One of the latest developments is Axon’s “Draft One,” an AI-powered reporting tool designed to create police report narratives from body-worn camera recordings.

Supporters argue that the technology will save officers time and reduce administrative burdens. Critics, however, raise important questions about accuracy, fairness, privacy, disclosure obligations, and the impact of AI-generated reports on the criminal justice system.

As criminal defence lawyers, we believe these issues deserve careful scrutiny.

What Is Draft One?

Draft One is an AI tool developed by Axon, the company that manufactures body-worn cameras used by many police agencies. The software generates draft police narratives based on audio captured through body-worn cameras.

According to Axon, officers must review, edit, and approve the draft before it becomes part of the official police report. The company promotes the technology as a way to allow officers to spend less time writing reports and more time policing.

While increased efficiency may be attractive to police services, efficiency alone cannot come at the expense of fairness, reliability, or accountability.

Why Does This Matter?

Police reports often play a significant role in criminal investigations and prosecutions. They can influence charging decisions, bail hearings, Crown screening, plea negotiations, and trial preparation.

When a report is drafted with the assistance of artificial intelligence, several important legal questions arise:

• How accurate is the AI-generated narrative?
• What information did the AI rely upon?
• Were any errors, omissions, or assumptions introduced?
• What edits were made by the officer?
• Can the defence obtain records showing how the report was created and every draft?

These questions may become increasingly important as AI-assisted policing becomes more common. 

Disclosure Concerns

One of the most significant issues is disclosure.

In Canada, the Crown has a constitutional obligation to disclose all relevant fruits of the investigation to the defence. If an AI system is involved in creating a police report, defense counsel may seek disclosure of:

• The original body-worn camera recordings;
• Audio transcripts;
• AI-generated draft reports;
• Audit logs;
• Revision histories;
• Policies governing AI use;
• Officer training materials; 
• Documentation explaining how the software functions; and 
• The timing of the generation report and the officer’s computer generated notes. 

If AI-generated drafts or related records are not preserved, it may become difficult to determine whether the final report accurately reflects what occurred during the investigation.

The inability to examine how a report (and the impact the report had on the officer’s notes) was created could become a significant issue in future criminal cases.

Can AI Influence an Officer’s Memory?

Another concern involves what some commentators describe as “memory contamination.”

Traditionally, officers prepare reports based on their own observations and recollections. Those reports may later be used to refresh memory when testifying in court.

However, if an AI system generates a narrative before an officer independently records their recollection, there is a risk that the officer’s memory becomes influenced by the AI-generated version of events.

Over time, an officer may remember the report rather than the underlying incident itself.

This issue could become particularly important during cross-examination when officers are asked to distinguish between what they personally observed and what was generated by software.

Accuracy and Hallucinations

Like all generative AI systems, Draft One is not immune from mistakes.

Even when a transcript appears accurate, AI systems can misunderstand context, misidentify speakers, omit important information, or generate incorrect details. Technology companies often refer to these errors as “hallucinations.”

This raises a simple but important question: “How carefully are officers reviewing AI-generated reports before approving them?”

A polished narrative may appear reliable on its face, but courts will likely remain concerned about whether the officer independently verified every statement before adopting it as their own.

The Loss of Human Judgment and Nuance

Police reports often contain more than a transcript of spoken words.

Officers frequently include observations about:

• Demeanour;
• Physical behaviour;
• Environmental conditions;
• Signs of impairment;
• Emotional state;
• Officer safety concerns; and
• Other contextual details.

Many of these observations cannot be captured through audio alone.

Criminal investigations frequently involve complex human interactions that require judgment, discretion, and interpretation. 

Critics argue that AI-generated narratives may oversimplify events or miss important nuances that could be relevant to a court’s assessment of the evidence.

Privacy and Data Storage Concerns

Questions also remain regarding data storage and privacy.

Axon indicates that Draft One operates within its broader digital evidence ecosystem, including Axon Evidence, a cloud-based evidence management platform. However, important questions remain regarding:

• Where data is processed;
• Where information is stored;
• Whether information ever leaves Canada;
• What third-party service providers have access to the data; and
• How long records are retained.

Because Draft One was developed using Microsoft’s Azure OpenAI infrastructure, defence lawyers, privacy commissioners, and courts may increasingly seek greater transparency regarding data processing and data residency practices.

These issues are particularly important where highly sensitive personal information is involved.

Should AI Contributions Be Identifiable?

As AI becomes more involved in policing, transparency becomes increasingly important.

One question that may eventually require judicial or legislative consideration is whether AI-generated portions of a police report should be clearly distinguishable from content created or modified by an officer.

If artificial intelligence contributes to a report, should the defence be entitled to know precisely which portions originated from the software?

Many would argue that the answer should be yes.

Transparency promotes accountability and allows courts to properly assess the reliability of evidence.

How Might Courts Respond?

At this stage, there are more questions than answers.

Canadian courts have not yet fully addressed the evidentiary and procedural issues that AI-generated police reports may create. Future cases will likely examine:

• Disclosure obligations;
• Reliability concerns;
• Admissibility issues;
• Officer credibility;
• Privacy implications; and
• The impact of AI-generated narratives on trial fairness.

Courts may eventually require greater disclosure regarding AI-assisted report generation, particularly where the technology plays a significant role in documenting key evidence.

The Bottom Line

Artificial intelligence may help police services reduce administrative workloads, but criminal investigations are not merely administrative exercises.

Accuracy, fairness, transparency, and accountability remain fundamental principles of the Canadian justice system.

As AI-generated police reporting tools become more common, defence lawyers will need to carefully examine how these systems operate, what records are preserved, and whether an accused person’s right to make full answer and defence is affected.

The technology may be new, but the underlying principle is not: every accused person is entitled to challenge the evidence being used against them.