Editor’s note: The authors of this article are three high school students who worked with leaders from the Columbia (Missouri) Police Department to develop prototype AI tools addressing surveillance video review, 911 call analysis and report writing. Their project, Operation: Safer Communities, placed among the top five nationally in the Presidential AI Challenge and earned them an invitation to the White House. In this article, they share what the development process taught them about AI’s potential in policing, as well as the security, public trust and human oversight questions agencies must address before putting such technology into operational use.
By Aayan Behura, Aneesh Calyam and Aanya Shetty
We recently had the opportunity to travel to Washington, D.C., for the White House Presidential AI Challenge. We competed as high school students with our project, Operation: Safer Communities, placing among the top five projects nationally. As AI continues to find its place in modern policing, our project can help administrators understand how to use this emerging technology to benefit their departments.
The first steps: A project grounded in current gaps
In recent years, there has been no shortage of new AI tools designed to help officers work as efficiently as possible. As we began our project, we sought to understand where current tools fail to meet departments’ needs and how we could better address those gaps.
To gain this understanding, we reached out to Barbara Buffaloe, the mayor of Columbia, Missouri, our hometown. From there, we were connected with Assistant Chief Mark Fitzgerald of the Columbia Police Department (CPD). Through this partnership, we established three focus areas for our project: surveillance footage review, 911 call analysis and report writing.
To address these areas, we created three AI tools.
The first was a video analyzer that uses two neural networks to identify anomalies in CCTV footage. The networks are trained on a crime dataset published by the University of Central Florida to identify common crimes, such as vandalism, break-ins and gun violence. Once an anomaly is detected, the tool reports it to officers for further investigation.
The second was a 911 call categorization tool that provides useful analytics, such as crime trends, crime hot spots, staffing patterns and assessments of 911 call operators for training purposes, to help administrators make critical decisions.
The final tool was an incident report generator that uses the Google Gemini API to organize officers’ unstructured notes, including the date, time, incident type and key facts, into draft police reports, potentially reducing the time officers spend behind a desk.
While developing the tools, we received constant feedback from officers and administrators at CPD, as well as officials citywide. We also consulted law enforcement leaders and public officials from coast to coast to gather national insights and optimize our tools.
Lessons from the field
To provide a roadmap for other departments navigating this landscape, we recognize that our journey was far from flawless. While building, sharing and exploring the potential implementation of our tools, we encountered several issues. These challenges can be classified into two broad categories: security and public trust.
First, using free resources to build our tools created a major data security issue. The backend for all our platforms was powered by Supabase, a resource designed for small, web-based projects. While precautions were taken to keep individual officers’ accounts private, files and information entered into our tools were stored in databases with minimal security protections.
Additionally, one of our tools, the report writer, was powered by the Google Gemini API. Because entering sensitive information into a large language model without appropriate privacy and security protections can put that data at risk, this presented an even more serious concern.
Administrators can avoid such issues before tools are even built. One major takeaway from our project was the importance of using secure, agency-approved systems and educating officers about the risks of entering sensitive information into public-facing platforms.
Another issue we have witnessed over the past few months is the challenge of earning the public’s trust. As we progressed through the different stages of the competition, our work repeatedly attracted attention through Facebook posts and news articles. Many people used the comment sections to share their feelings about the use of AI in policing.
Specifically, we saw two main concerns among community members: fear of excessive surveillance and broader concerns about AI, including its environmental impact. These reactions reflected growing concerns beyond our town.
Nationwide, departments need to emphasize that the purpose of AI is not to invade individuals’ privacy but to help officers respond to crime more efficiently and administrators organize their departments to better protect the public. For example, emphasizing that our AI tools only help analyze data that is already being collected could alleviate some of the concerns we encountered.
While we have yet to speak directly with the public about our tools, we encourage departments looking to implement new technologies to hold community conversations to raise awareness, address common concerns and misconceptions, and build public trust before deployment.
Simply welcoming these conversations can be meaningful. After seeing the public response to our work, command staff at the local police department and city officials have said they feel better prepared to navigate the emergence of new technologies.
Our project as a blueprint for responsible AI adoption
Throughout our conversations with law enforcement experts, we identified one common point of agreement: The optimal role of AI in policing is to augment, rather than replace, officers and administrators.
Seeing this consensus, we centered our tools around a human-in-the-loop design, ensuring that decision-making power never leaves the hands of human officers. For example, we emphasized that our tool for detecting anomalies in surveillance footage should flag timestamps for human review, not make the final determination about potential bad actors.
Similarly, our incident report generation tool has built-in checks that require an officer to review the document completely before it can be downloaded or copied. Departments that prioritize retaining human judgment when designing tools can turn AI into an asset rather than a replacement.
Beyond the platforms themselves, our project helps keep human administrators in control by enabling them to adapt tools to meet their departments’ unique operational challenges. Using emerging AI technologies, individual departments now have an unprecedented ability to create custom tools to meet their needs.
Our work demonstrates that departments large and small can explore this opportunity. Using free resources and APIs, we were able to build working prototypes in just a few weeks based on needs identified by local law enforcement officials.
If three high school students working in a basement over winter break can achieve this, seasoned law enforcement professionals have an even greater ability to build custom tools that fit their precise operational needs.
The bottom line
When we first began this project, we thought the technology would be the hard part. However, with AI’s growing capabilities, the bottleneck turned out to be something else: balancing the desire to improve departmental efficiency with the need to keep human officers at the forefront of every decision.
Finding that balance required discussions within our community and with local officials and law enforcement administrators across the nation. Departments that avoid these conversations will struggle to implement any AI tool, no matter how technologically capable it may be.
About the authors
Aayan Behura, Aneesh Calyam and Aanya Shetty are students at Rock Bridge High School in Columbia, Missouri, with a passion for technology and AI. They have worked on several projects focused on using AI for the good of society. Most recently, they collaborated with the Columbia Police Department to build three AI-powered tools designed to improve administrative and investigative efficiency. They also shared their tools with top officials at several large departments across the nation and were invited to the White House for their work.
