Artificial intelligence is transforming numerous administrative processes and is also beginning to occupy a place in institutions responsible for ensuring public safety. In the United States, various police departments are exploring tools capable of transcribing conversations, organizing information, and preparing drafts of official documents. In this context, the San Diego Police Department in California is studying the possibility of using artificial intelligence to assist its officers in drafting police reports. The initiative presents opportunities for efficiency, but also raises questions about accuracy, privacy, and judicial safeguards.
According to information published by Infobae on October 14, 2026, the project contemplates a possible contract worth between $500,000 and $3 million. This figure corresponds to an estimate linked to the initiative and does not demonstrate that the department has awarded a contract for that amount. The proposal is in an exploratory phase, so institutional intentions must still be distinguished from final decisions. Its analysis makes it possible to examine how authorities can take advantage of technology without assuming that benefits requiring specific evaluations have already been demonstrated.
San Diego Studies a New Technological Tool
On July 31, 2026, the San Diego Police Department issued a request for information to learn about available solutions that use artificial intelligence in the preparation of police reports. The objective is to study tools that make it possible to improve drafting efficiency and reduce the time officers spend on administrative tasks. According to media coverage, the initiative's estimated budget ranges from $500,000 to $3 million. However, a request for information is not equivalent to a contract award and does not confirm that a provider has been selected.
Ashley Nicholes, a department spokesperson, explained that the institution was exploring existing technological options and recommended practices for using them. She also indicated that, if the project moved forward, elected representatives and community members would be consulted. These statements indicate that the initiative still requires institutional evaluation. Therefore, it would not be correct to claim that all San Diego officers are already using artificial intelligence to draft their reports or that the purchase has been definitively approved.
The department considers that the technology could help reduce the administrative burden and allow officers to devote more time to responding to service calls. Officials must document interventions, record statements, and describe relevant circumstances surrounding incidents. These tasks are essential for maintaining official records and facilitating subsequent investigations. Nevertheless, the benefit of automating them will depend on the accuracy of the results and the time required to review them.
How Does Artificial Intelligence Applied to Reports Work?
Artificial intelligence systems designed to draft police reports can combine speech recognition, automatic transcription, and generative language models. In general terms, they process information from recordings, transcripts, or officers' notes and produce an organized draft. The officer can review the content, correct errors, and complete the necessary information before submitting the official document. The purpose is to automate certain writing tasks, not to eliminate the professional responsibility of the person who signs the report.
One tool known in this field is Draft One, developed by Axon, a company that markets body cameras and other technologies for public safety agencies. The system uses audio from body cameras to help generate narrative drafts. Its existence demonstrates that commercial products designed to support this activity are available, but it does not confirm that San Diego has chosen this solution. Any tool the department might ultimately adopt would have to be evaluated according to its features, contractual conditions, and results obtained in relevant testing.
It is important to distinguish a transcription from a police report. The former attempts to reproduce the words captured in a recording, while the latter organizes events and describes the information considered relevant. Generative artificial intelligence can produce a coherent narrative, but the fluency of the text does not guarantee that every statement is correct. If a recording contains noise, overlapping voices, or ambiguous expressions, the system may misinterpret a phrase and transfer that error to the document. Therefore, the quality of the report depends on the original sources, the technological processing, and human review.
Potential Time Savings
Reducing administrative tasks is one of the main arguments in favor of artificial intelligence in policing. Officers must complete documentation after responding to incidents of varying complexity, which can limit the time available for other duties. If a system automates repetitive tasks without compromising the quality of records, it could improve work organization. It could also help standardize the format of certain documents.
Chula Vista, a city near San Diego, provides a point of comparison because its police department already uses artificial intelligence tools in certain activities. The department's chief, Roxana Kennedy, has indicated that officers can spend between three and five hours per shift on tasks related to reports, according to statements reported by the media. This figure should be understood as an estimate attributed to the official, not as an average applicable to all agencies. The documentation workload varies according to the type of incident, internal procedures, and administrative requirements.
Actual time savings must be measured by taking the entire process into account. If the system generates a draft quickly but the officer needs a great deal of time to verify statements and correct errors, the net benefit could be limited. Therefore, a rigorous evaluation must account for both the time required to generate the document and the time spent reviewing, modifying, and approving each report. Only a comparison with traditional methods will make it possible to determine whether artificial intelligence actually improves productivity.
What Does Research Indicate About Accuracy?
Accuracy is one of the main concerns associated with the automation of police reports. Ian Adams, a former police officer and criminology researcher, has studied the use of these tools in police work. A study published in the Journal of Experimental Criminology examined the Draft One system through a controlled trial involving 85 officers and 755 reports. According to the results described in coverage of the research, no statistically significant reduction in drafting time attributable to the system was found.
This result challenges the idea that automating writing necessarily produces immediate time savings. Adams has also pointed out accuracy problems in evaluations of reports prepared with the assistance of artificial intelligence. These observations deserve attention because the documents may be examined by prosecutors, defense attorneys, and judges. However, the results obtained with one specific tool do not demonstrate that all available systems operate in the same way.
Technology changes, and new versions may incorporate improvements in audio processing and text generation. For that reason, evaluations must correspond to the product intended for use and to actual working conditions. Departments should not assume that a tool is effective simply because the provider promises to save time. Artificial intelligence must undergo verifiable testing that makes it possible to measure its accuracy and operating costs.
Errors and Consequences for Justice
A police report can help reconstruct an incident, guide an investigation, or facilitate subsequent decisions by the authorities. Although it does not determine a person's guilt on its own, its content can influence the interpretation of events. An error in the description of a conversation or an action can become significant when the document is used during judicial proceedings. Therefore, the incorporation of artificial intelligence requires verification that the report reflects the available evidence.
Potential risks include incorrect transcriptions, statements attributed to the wrong person, and the omission of relevant details. A narrative alteration may also occur when the system presents a connection between events as certain even though the recording does not establish it. These risks do not mean that all automatically generated documents contain errors. They indicate that convincing writing does not constitute a guarantee of truthfulness.
Andrew Ferguson, a law professor at George Washington University who specializes in police surveillance, has warned about the danger that algorithmic modifications may affect the reconstruction of facts in criminal proceedings. His concern relates to the possibility that certain expressions may subtly change the meaning of an action or statement. Nevertheless, manually prepared reports can also contain errors or subjective interpretations. The challenge is to identify the specific risks of automation and establish controls that make it possible to detect and correct them.
Human Oversight and Professional Responsibility
Human review is an essential safeguard when artificial intelligence is involved in preparing official documents. The officer must verify that the facts described correspond to the available sources and that no unsupported details have been added. Responsibility cannot be transferred entirely to the technology provider or disappear because a tool generated the text. Approval of the report must represent an effective review of its content.
The California Commission on Peace Officer Standards and Training, known as POST, has published model guidance related to the use of artificial intelligence in administrative documents. This guidance includes human review and approval of the results. It is important to distinguish recommendations or model policies from obligations established by law. Each department must ensure that its internal procedures comply with the applicable legal provisions.
Oversight also requires training. Officers need to understand the system's limitations, recognize common errors, and know when they must compare the text with the original recording. If review becomes a formality, there is a risk of approving incorrect documents because of excessive trust in a well-written narrative. Therefore, procedures should define which aspects must be checked, how corrections are recorded, and who authorizes the final version. Artificial intelligence must function as a support tool, not as an automatic source of truth.
Privacy and Public Trust
Police recordings may contain sensitive information about victims, witnesses, individuals under investigation, and citizens who are not involved in a crime. They may also reveal voices, images, home addresses, and details about emergencies or family situations. When an artificial intelligence tool processes these materials, it is necessary to know where they are stored, who can access them, and how long they are retained. Contracts must define access conditions, security measures, and procedures for responding to potential incidents.
Khalid Alexander, a professor at San Diego City College and founder of the community organization Pillars of the Community, has expressed opposition to the proposal. His criticisms include concerns about the dehumanization of police work, data security, and excessive trust in automated systems. These statements represent a critical position within the debate, but they do not demonstrate that the tool San Diego might contract will necessarily produce those results. They do, however, show the importance of listening to the community before introducing changes to police documentation.
Public trust also depends on institutions explaining when they use artificial intelligence and how they review generated documents. If a recording contradicts a report, it must be possible to identify the source of the error and determine what corrections were made. Retaining drafts and maintaining audit logs can facilitate this analysis. However, the existence of records does not guarantee that every problem will be detected. Transparency, access controls, and effective review procedures must be combined.
The Experiences of Los Angeles and Chula Vista
San Diego is not the only city incorporating or studying automated tools to support police work. The Los Angeles Police Department has included Axon technologies in contractual agreements whose total value has been described by the press as approximately $235 million. That agreement covers various products and services, including features related to preparing summaries from body camera audio. Therefore, it should not be interpreted as though the entire amount corresponded exclusively to an automated report-writing program.
Questions have also been raised about potential artificial intelligence biases in Los Angeles. Dean Gialamas, the department's chief technology officer, indicated that the systems being used were non-agentic, meaning they were not designed to make decisions or perform tasks autonomously. Officers were required to review the documents before submitting them. This distinction makes it possible to differentiate a program that generates drafts from one that acts on its own, although human oversight does not eliminate every possible error.
The experiences of other cities can help San Diego identify practical problems and evaluation criteria. However, the results cannot automatically be transferred from one jurisdiction to another because procedures, workloads, and operating conditions differ. The department needs to verify whether a specific tool meets its needs and complies with applicable regulations. External experience can guide the decision, but it does not replace local testing.
How Should Implementation Be Evaluated?
Before contracting a system, San Diego should define indicators that make it possible to compare the results with conventional methods. These may include the total time spent on each report, the number of corrections, and the frequency of significant errors. It is also important to verify whether the system correctly attributes statements and preserves the essential details of each incident. Establishing these criteria before testing helps prevent the evaluation from being limited to highlighting favorable results.
A pilot program would allow artificial intelligence to be tested under real-world conditions before its use is expanded. Evaluators could compare drafts with the original recordings and other reliable records, distinguishing stylistic errors from alterations that change the meaning of events. It would also be advisable to test performance across different types of incidents and audio conditions. The results should be documented to identify limitations and determine which additional controls would be necessary.
The economic analysis must consider more than the initial contract price. Licenses, technological integration, storage, training, maintenance, and auditing can generate additional expenses. If the system produces drafts quickly but requires extensive reviews, the actual savings could be lower than expected. Conversely, if testing demonstrates sustained improvements without reducing accuracy, the investment could be justified. The decision must be based on verifiable results and a transparent comparison of costs and benefits.
Finally, responsible implementation requires clear rules regarding approval, draft retention, and data protection. The participation of legal experts, officers, and community representatives can help identify difficulties before the program is expanded. It is also advisable to communicate evaluation results and explain how errors will be investigated. These measures do not eliminate all risks, but they make it possible to make better-informed decisions and adjust procedures to the available evidence.
The evaluation of artificial intelligence for drafting police reports in San Diego reflects the challenge of incorporating new technologies without compromising the safeguards of the justice system. The proposal seeks to identify tools that may reduce the administrative burden and improve the organization of police work. However, the request for information and the estimate of between $500,000 and $3 million do not amount to a definitive contract. Any decision must consider both the potential benefits and the documented limitations.
The available evidence does not make it possible to claim that all artificial intelligence systems are ineffective or that they always produce less accurate reports than those prepared manually. It does, however, justify requiring rigorous testing, human oversight, transparency, and procedures that make it possible to track the changes made to each document. California law establishes obligations to identify the use of these tools, retain initial drafts, and maintain audit logs. These measures provide a foundation for accountability, although they do not replace technical evaluation or the protection of personal data.
Ultimately, the value of artificial intelligence in policing will depend on its ability to improve procedures without weakening the reliability of information. Automation will only be justifiable if its benefits are demonstrated and officials retain responsibility for official documents. Accuracy, privacy, and transparency must remain priorities throughout the entire process. For organizations interested in integrating technology strategically, ITD Consulting invites you to learn about its consulting services and technological solutions. To request information and explore options tailored to your organization's needs, you can write to [email protected].
EN 
ES