Words on a blue flag written in a bizarre language. A child’s legless body melds with a bench. Five men of different sizes, all wearing the same purple polo shirt, stare off at the mountain landscape.

This surreal image is a concept design for a monument commemorating the United States’ 250th anniversary, planned for installation somewhere in Heber City this year. The concept design was generated by city staff using artificial intelligence. 

In recent years, the use of generative AI has become increasingly common across workplaces and areas of life, including small-town government. That’s why the Heber City Council adopted a policy guiding the use of AI by city employees.

The approach is neither to encourage nor discourage AI use, but to set guardrails for its inevitable use. 

During an initial discussion about the policy in January, City Councilor Aaron Cheatwood said, “Rather than restricting, teaching and explaining and understanding is way more powerful.”

That’s why all employees will be required to complete quarterly, informative trainings about AI through Executech, the business providing IT support for Heber City.

AI use will be approved by department heads and monitored by the city’s IT director, Anthon Beales.

The use of free AI tools is restricted. That’s because free AI tools generally have fewer privacy protections, explained Executive Assistant to the City Manager Lainee Meyers, who wrote the policy.

“They often store and often reuse the data that is entered into them. That can most certainly create an unintentional risk of sharing sensitive information,” she said.

For that same reason, city staff are not allowed to enter confidential, protected or personally identifiable information — such as addresses, names and phone numbers — into any AI tool unless authorized by a department head, or if the tool is explicitly approved for such use.

The Heber City Police Department currently uses two AI tools that handle sensitive information. 

The first is Code Four, which generates first drafts of police reports based on body cam video and audio, which the Police Department purchased in January following a two-month free trial period.

The second is Blue Voice, an AI assistant that searches from any data the Police Department enters into it, which includes its policy manual, city code, police officers’ phone numbers and addresses and gate codes for areas like the animal shelter and gated communities.

Both programs are compliant with the FBI’s Criminal Justice Information Systems Security Policy.

Other than Code Four and Blue Voice, the only other AI software currently used by city employees is the Microsoft Copilot assistant. Information entered into Microsoft Copilot is documented within a local government cloud. 

If AI is used to contribute to materials intended for elected officials or the public — including public notices, staff reports and web content — a disclosure including what software was used is required.

In that case, all drafts, prompts and outputs that resulted in the product are subject to records retention, meaning a member of the public could submit a Government Records Access and Management Act (GRAMA) request to access those records.

Some uses of AI are prohibited per the policy, including using AI to make a final decision related to hiring, termination or other staffing concerns, or to determine eligibility for public services or benefits.

The policy emphasizes that final decisions must be made by humans rather than AI and that humans are ultimately responsible for any products or decisions made with the assistance of AI.

Heber City is not the first municipality in Wasatch County to consider an AI policy. Midway City Attorney Corbin Gordon proposed a similar policy to the City Council last month, which the city is still drafting. 

The Wasatch County government has not considered a similar policy. However, County Councilor Erik Rowland used AI last month to analyze the Utah Department of Transportation’s draft environmental impact statement for the Heber Valley bypass. That AI analysis served as inspiration for a letter he drafted to UDOT criticizing its proposed route. Elected officials from across Wasatch County signed it.

While Heber City’s AI policy does not explicitly encourage or discourage use of the technology, Cheatwood believes AI could have untapped benefits for city staff.

“It saves tons of time,” he said. “I think internally, we can encourage (usage) the right way. We know that every (department) says (they’re) understaffed. What can you give up?”

The post Heber City adopts policy guiding AI use by staff appeared first on Park Record.

The Heber City Police Department went viral recently after Chief Parker Sever told a story he’d heard from an officer at West Jordan Police Department about an AI program that generates police reports based on body camera audio. After overhearing a movie playing in the background, it generated a report claiming that a police officer had transformed into a frog.

The story was picked up by Forbes, Vice and Axios, all originating from The Park Record’s reporting and a follow-up story by Fox 13. But the attention hasn’t stopped Sever from implementing such technology into the Police Department’s practices. 

The Police Department purchased Code Four, an AI software that drafts police reports based on body camera footage, following a pilot program that pitted the startup against law enforcement technology juggernaut Axon’s competitor, Draft One. The Police Department engaged in two-month free trials for each program, which ended in January.

Sever was interested in the software for the time savings, which he said would allow officers more time on the streets instead of in the office.

Sever said one reason Code Four was chosen was because of its greater capabilities. Draft One creates reports based only on body camera audio, while Code Four analyzes video as well. That allows for greater detail in the reports. 

For example, the software can identify clearly visible objects, like a laptop, and provide information like brand and model, distinguishing features, serial numbers and a typical value range for the object.

Code Four co-founder George Cheng explained that the software is compliant with the FBI’s Criminal Justice Information Services Security Policy, including end-to-end encryption. Additionally, the software uses Customer-Managed Encryption Keys, meaning the Heber City Police Department has the “ultimate authority over who can access the video and audio data” collected from body cameras.

Code Four is available to officers who have completed their probationary period. 

Police reports are signed off by the officer who created them and their supervisor before being sent to the records department. Sever believes this provides a sufficient defense against inaccurate reports. He also said officers who were caught not reviewing their AI-drafted reports would be disciplined or terminated.

Code Four costs the department about $6,000 to $8,000 annually, while Draft One would have cost around $30,000 annually.

In a previous article by The Park Record, David Ferguson, executive director of the Utah Association of Criminal Defense Lawyers, raised concerns about AI software writing police reports. He said it would be unethical for defense attorneys to trust an AI’s interpretations as fact, meaning attorneys would increasingly be forced to review body camera footage themselves.

Sever does not share Ferguson’s concerns. 

“I actually think they needed to look at the cases closer when the officers were writing them,” he said. “If you ask my records clerks right now, our reports are better.”

He attributed that to the greater level of detail in a report drafted by Code Four instead of an officer. He said officers tend to provide only the essential facts because of time constraints an AI program is not subject to.

Sever has since identified a second AI program for the Heber City Police Department. It’s an AI assistant for police officers called Blue Voice.

Blue Voice works similarly to ChatGPT or any other AI search engine and is accessible from cell phones, tablets, computers and mobile data terminals. The difference is that it only searches from data and documents uploaded to it by the police department that uses it. 

The Heber City Police Department has uploaded its policy manual, maps of areas like the public schools and the codes for Heber City, Midway, Wasatch County and the state of Utah, all of which are public sources of information.

“You’re just limited by your imagination,” Sever said.

The Police Department has also uploaded private information, like the phone numbers and addresses of all officers and the gate codes for areas like the animal shelter and gated communities.

Blue Voice is compliant with the FBI’s Criminal Justice Information Systems Security Policy. 

Administrator accounts like Sever’s are required to use multi-factor authentication. All data uploaded by law enforcement agencies is end-to-end encrypted. And on Blue Voice’s end, all employees who access, process or maintain sensitive data are fingerprinted U.S. citizens who have undergone thorough background checks and information security training and signed non-disclosure agreements, according to the software’s website.

Blue Voice was founded by a group of former law enforcement employees and engineering professionals from the East Coast. 

“Immediate access to vital information is critically important for a number of reasons,” said Michael Gropman, Blue Voice’s chief business officer. “It improves safety and security for the officers and the public, decreases mistakes and ensures all citizens’ civil liberties are acknowledged and respected.”

The software has been implemented in about 250 police departments and sheriff’s offices in 14 states, according to Gropman. The Heber City Police Department is the first law enforcement agency in Utah to use it.

Sever imagines Blue Voice being used to help sift through code and policy documents more easily. 

That could be especially helpful when an officer encounters an unfamiliar situation.

“There’s a lot of little nuances to code that the officers don’t have memorized because they just don’t encounter it very often,” Sever said.

He provided an example by demonstrating a search for “water theft.” None of the documents uploaded use the phrase “water theft,” but Blue Voice was still able to understand the search and outline unlawful or unauthorized uses of water.

Sever described the policy manual as the lifeblood of the Police Department, governing all practices and interactions with the public. 

“Just about everything you can possibly think of, there’s a police policy for it,” he said. 

That includes policies on public breastfeeding, hate crimes, aircraft incidents and more.

Sever said his frustration with holding officers accountable to so many dense policies and codes that are difficult to access is what led to implementing Blue Voice.

The department got a “smoking deal” of a free year, which begins this month. After that year is up, Blue Voice will cost the Police Department about $7,000 per year. 

Reactions to the Heber City Police Department’s implementation of AI have been positive both internally and externally, Sever said, adding that he has received multiple calls from law enforcement agencies across Utah asking about Code Four.

The Heber City Police Department is currently the only law enforcement agency in the Wasatch Back to have implemented AI into its policing.

The post Heber City Police Department continues to be early adopter of AI appeared first on Park Record.

A teacher from 1926 could never have predicted a classroom full of calculators, cell phones and laptops.

It’s only been a hundred years — a relatively short period of time for human civilization — but how people interact with each other and the world has changed dramatically. For decades now, teachers have had to be vigilant, keeping an eye out for students looking up test answers on their phones or copying a Wikipedia article into their essays.

But what do school administrators do when artificial intelligence, the latest technological advancement taking the country by storm, gives students and teachers the ability to create, submit and publish work that has never been touched by a human?

The Park City School District last year established a committee with representatives from each school to tackle that very question. Since then, the district has constantly revised and updated its approach to artificial intelligence in the classroom, including a school-sanctioned AI program to be used as a teaching tool.

Ryan Garey, a technology specialist for the district, presented information to the school board on Tuesday regarding the district’s current approach to AI in schools. He said the goal is to use AI ethically and intentionally to support personalized learning. However, he also emphasized the importance of academic integrity and said the district’s experimentation with AI will help students learn how to integrate evolving technology into their lives as a core skill.

Students were never allowed to access ChatGPT on campus, and teachers were prohibited from the platform last year when the AI system started incorporating explicit content. But Garey said a new AI system designed specifically for educators, called MagicSchool, was made available for both students and teachers in August.

MagicSchool differs from other AI programs because it relies solely on parameters established by teachers. Students can’t log in and use MagicSchool whenever they want. Instead, teachers need to create standards and input information to guide students on how to use the AI tool.

In many cases, MagicSchool simply provides students with personalized feedback on their assignments at a faster rate than a teacher could by hand.

Garey said MagicSchool was also a good program for the district because it doesn’t share data or analyze inputs. AI systems like ChatGPT constantly monitor the conversation between the AI and its user to teach and improve the system. MagicSchool, however, is “static,” meaning it does not retain any of the information entered into its system, including teacher instructions.

Garey added that the rollout of MagicSchool was done slowly and intentionally to ensure it functioned how it was supposed to and that there were guidelines in place for how to use it. The pilot phase was from August to December, and 108 teachers were using MagicSchool as a teaching tool by winter break. It is up to each teacher’s discretion as to whether they want to use AI in the classroom.

Another 342 Park City staff members are using MagicSchool’s “teacher tools” outside of class.

Teachers are also allowed to use other AI systems, including Google Gemini, Canva, Adobe Express, Diffit and Nearpod. Diffit and Nearpod are AI tools specifically marketed to educators to create lesson plans, search for instructional materials and format classes to be more interactive.

“We’re continuing to remind our staff that it is a tool, not the tool,” said Superintendent Lyndsay Huntsman. “Think about calculators, laptops, now AI. It’s a tool, but it should never replace that human interaction and the relationship piece that our teachers are so good at.”

Garey said the district plans to roll out MagicSchool to its elementary schools soon, but it will mostly be a tool for teachers. AI will be restricted in classrooms for students until fourth grade.

The post Park City School District uses school-sanctioned AI program as teaching tool appeared first on Park Record.

The Heber City Police Department is test piloting two AI software programs that generate police reports based on body cam footage: Code Four and Draft One by Axon. In November, Sgt. Josh Weishar invited Park Record Heber City reporter Cannon Taylor to test Draft One.

I knew from working with Otter, a transcription software used by journalists, that AI audio transcription can be riddled with errors. For example, “Heber” is often transcribed as “Hebrew” in my interview recordings. Because of Otter’s capacity to make errors, I always go back into the audio recording itself and verify that each quote I use is accurate. 

Knowing that Draft One works based on body cam audio, I thought the software could be tricked into writing a faulty report. 

Using what I knew about AI transcription software, I fabricated a fictional incident I would be testifying about to Weishar that included some tricky turns of phrase, slang terms and double meanings that might confuse Draft One.

The premise began with me alone in my second-floor apartment at 1 p.m., when I heard a “rat-a-tat” from the street outside. I ran to the window to look outside and saw a “smoking chick with a gun” pointing it at a storage unit across the street with both hands.

I wondered if Draft One would assume that I was describing the woman as attractive. Maybe it would think the woman was smoking a cigarette despite holding the gun with both hands. There was another possibility that it would think I was referring to a smoking gun. 

I exclusively used the word “chick” to refer to the woman because of her appearance. I described her as bare, except for yellow paint and feathers all over her body. It was a long shot, but there was a chance the bizarre getup would make the software think I was talking about a baby chicken.

I testified that I saw the woman “steel herself against a red truck” before moving towards the unit across the street. It was an incorrect usage of the phrase, but I was wondering if the AI might think “steel” referred to theft.

I then said I saw a “dark knight” outside my apartment. The woman shot him, and “the knight fell.” 

Yes, I was referring to Batman, but I thought the AI might think I meant the evening sky, even though the imaginary incident took place in the early afternoon. 

Batman got back up, unscathed, and threw, “like, a bat. But it also might have been a boomerang.”

The confusing description of the superhero’s signature batarang could have been taken any number of ways: that Batman threw a bat (animal) or a bat (sports equipment), or maybe even that he threw in a manner similar to the winged mammal. 

I concluded by saying that the object Batman threw knocked the gun onto the ground. He then kicked the woman and tossed her into the storage unit, locking it and disappearing.

I mentioned that I had been “kicking back with Mary Jane” the entire time, so I couldn’t entirely trust what I saw. It was a reference to marijuana, but there was a chance Draft One would think there was another person with me in the apartment.

I assumed my fictional incident was baffling enough that at least one error would make it into the report. The bits about the knight falling and the woman steeling herself against a truck, I thought, were most likely to confuse the AI.

After entering the police station, Weishar took me into a meeting room, where I reported the fabricated crime to him as an Axon body camera secured to his torso recorded the conversation. 

Weishar seemed bemused by the crime I was reporting. After all, the suggestion he’d provided of an incident we could run through was a lost cat, not Batman apprehending a chicken supervillain.

Weishar provided commentary for the body camera, which, he explained, helps in generating an accurate and detailed report.

The first time we ran through the incident, we used a backup body cam as Weishar’s was charging. 

But when we went into Weishar’s office to access the recording and report, they couldn’t be found in the database.

Weishar explained that because the backup body camera was not assigned to him, he didn’t have access to the recording.

We role-played the scenario again, now using Weishar’s body camera.

Because it was the second time running through the same story, Weishar seemed to notice my repetition of particular phrases — like how the woman “steeled herself against a red truck” — and asked a few follow-up questions he hadn’t asked during the first scenario. For example, when he asked what I meant when I said the woman “steeled herself,” I explained that she was doing so emotionally.

The second recording was uploaded and a police report was generated in a matter of minutes. 

Every detail was accurate to the conversation I had with Weishar; none of my tricks had made it into the report.

If any detail of the report was inaccurate, it would be up to the officer to correct any mistakes before the report was submitted. Police reports are reviewed by a supervisor and the records department.

The report adds prompts throughout the report where the officer can provide additional detail about certain elements of the incident, like the suspect or the location. These prompts must be deleted, though not necessarily answered, to submit the report. 

I was surprised by the accuracy of the report, given my firsthand experience with the pitfalls of AI transcription software. 

I wondered if a police report generated from the first run-through of the scenario would have turned out differently, given the differences in my interactions with Weishar between the two scenarios.

I also couldn’t help but wonder how Draft One would respond to the litany of variables that could emerge during an actual incident. For example, Weishar and I had used Draft One in a quiet room, but what if we had used it in a noisy environment — like a windy day on Highway 40? At least in Otter’s case, the software struggles to pick up voices in environments like that.

Finally, Weishar’s asking follow-up questions and providing narration for the body camera showed me that the accuracy of Draft One isn’t the sole burden of the software, but the officer as well.

The post A test of Heber City Police Department’s AI software appeared first on Park Record.

Heber City Police Chief Parker Sever appeared before the City Council in late October and informed them an officer had shape-shifted into a frog.

At least, that’s what he heard from his son-in-law, who works for the West Jordan Police Department. That department had recently implemented artificial intelligence software that generates police reports using body camera footage. The details aren’t always 100% accurate — which is why officers review them.

“I read the report, and I’m like, ‘Man, this really looks like an officer wrote it,’” Sever recalled. “But when it got to one part, it said, ‘And then the officer turned into a frog, and a magic book appeared and began granting wishes.’ … It was because they had, like, ‘Harry Potter’ on in the background. So it picked up the noise from the TV and added it to the report.”

The Heber City Police Department is test piloting similar software. 

The candidates are Code Four, a startup launched by two 19-year-old MIT dropouts earlier this year, and law enforcement technology juggernaut Axon’s Draft One. Select weekday officers have access to Draft One, while weekend officers use Code Four. 

The two-month free trial periods for these programs expire in January.

The selling point of an AI software that creates police reports is the time savings. 

Axon and Code Four cite similar statistics that police officers spend up to 40% of their time writing police reports.

 “I think most of the public wants our officers out on the street, not in the office,” Sever said.

That ‘brand new car’

Josh Weishar, a Police Department sergeant who had exclusively used Draft One, testified to the time savings on Nov. 10, about a week into the pilot program.

Weishar thought the AI-generated police reports were typically more detailed than what he would have written.

“An officer’s going to go with the basic information,” he said. “I think you’ll get more info, or at least, definitely more detailed info, than you would from just typing it out on your own because you’re just going to remember the key facts instead of all the detail that comes from that AI’s portion.”

That was one of the reasons the software appealed to Heber City Police — an officer might not be as detailed if they’re “down 10 reports,” Sever said, but AI can always include the same level of detail.

Both Draft One and Code Four allow the user to customize the level of detail the AI software includes in the report, from basic synopses to in-depth play-by-plays, and can generate reports from interactions in English and Spanish.

The major difference between the software is that Draft One creates a report based on bodycam audio, while Code Four uses bodycam audio and video.

End-of-shift reports, which provide an overview of incidents over an entire shift, will continue to be written exclusively by humans.

“Most of our officers are kind of awestruck with (Draft One) because it’s such a new, innovative thing for us,” Weishar said. “It’s like that brand new car that’s got all the features to it. For us, it’s crazy that you can just press a button and it’ll tell you everything about the case that you were on and give you a pretty decent police report to edit.”

How Draft One works

In a demonstration of Draft One, Weishar showed that the draft report is littered with prompts that ask the officer to provide additional detail about certain elements, like what a suspect looked like or the state of the crime scene. The prompts, which have to be deleted or edited to continue, are intended to force the officer to review the report with a fine-tooth comb. 

Once they’ve completed their edits and approved the report, the officer copies and pastes the report into the Police Department’s database and submits it as they would any other police report.

The original report generated by Draft One is not retained. However, the software creates an “unalterable digital audit trail” that captures who used the tool, when they used it and what evidence was involved, according to Axon’s website.

The report ends with a disclosure that is turned on by default: “I acknowledge this report was generated using Draft One by Axon. I further acknowledge that I have reviewed the report in detail, made any necessary edits, and believe it to be an accurate representation of my recollection of the reported events. If needed, I am willing to testify to the accuracy of this report.”

The disclosure can be customized or turned off.

Comparison shopping

If the Police Department chooses to continue with Draft One, it would cost about $30,000 of taxpayer dollars per year, according to Sever. 

Code Four co-founder George Cheng was critical of Axon because of additional add-ons, such as extra video storage and processing. He estimated it costs up to $400 per officer per month because of these add-ons. Code Four is significantly cheaper at $30 per officer per month. Sever estimated it would cost Heber City between $6,000 and $8,000 annually. 

Cheng visited the Taylorsville and Heber City police departments before Thanksgiving to demonstrate how the software works. He’s made similar house calls at the Spanish Fork, South Jordan, Kaysville and Draper police departments.

Sever said he’s only able to authorize expenditures up to a certain amount without Heber City Council approval. Purchasing Draft One for around $30,000 per year would require such permission. Buying Code Four at around $6,000 to $8,000 per year would not.

As of December, Heber City Police were leaning toward Code Four because of its lower cost and ability to write more comprehensive reports.

Another case

The Heber City Police Department isn’t the only law enforcement agency in the Wasatch Back test-piloting AI software.

The Summit County Sheriff’s Office recently completed a 90-day pilot program of Draft One among six patrol deputies. The Park City Police Department and Wasatch County Sheriff’s Office have not implemented any AI software practices.

Sheriff’s Sgt. Skyler Talbot said his outlook on AI in policing is a balance of caution and optimism.

“It’s important that we recognize the capabilities of AI and the strong potential to increase operational efficiency,” he said. “I think it’s important that we stay on the cutting edge of that as it develops. But with that … I don’t want to say we should be skeptical, but we definitely need to be cautious.”

Talbot said Draft One was not used to draft reports for “major incidents or any incident that might result in felony level charges” during the pilot program.

On Nov. 25, Talbot informed The Park Record that the Summit County Sheriff’s Office had chosen not to continue with Draft One past the free trial.

He explained that Draft One has an administrative setting that forces the AI software to insert obvious errors into every report. 

“What that did is prompted and really forced the deputy to go through, read the report, check it for accuracy and make those corrections before it would allow them to continue,” Talbot said. Essentially, the time savings weren’t as impressive.

Having that setting on was a necessity if the Summit County Sheriff’s Office were to implement Draft One to ensure deputies were reading the draft report carefully and checking it for accuracy before submitting it, Talbot explained. 

Meanwhile, the Heber City Police Department does not have this setting turned on. Sever said he was not aware of the setting and would have to look into its value.

Talbot said it wasn’t “fiscally responsible” for the Summit County Sheriff’s Office to use taxpayer funding on the program, though he clarified that the stance could change in a matter of years or months as the field evolves.

“Again, this is useful. This is great technology for our department size, for our call volume,” he said. “It just makes more sense, at this point anyway, just to do things the way we were doing them before.”

Ethical concerns

Not all are optimistic about AI-drafted police reports. 

David Ferguson, executive director of the Utah Association of Criminal Defense Lawyers, detailed some of the potential pitfalls of such AI software. He said that police reports are vital to defense attorneys because they can reveal whether the writer has any biases. 

But because generative AI can “hallucinate,” getting details wrong or making them up entirely, it would be “unethical” for a defense attorney to take an AI-drafted police report at face value, leading them to review the bodycam footage to verify basic details in more and more cases, Ferguson said. That could increase workload for defense attorneys, particularly overworked public defenders. 

Prosecutors, on the other hand, typically rely on the evidence given to them by the police, including police reports.

A report generated by an AI software also has the potential to modify an officer’s memories of an incident. If an officer who used generative AI to draft a report testified in court, they could end up testifying based on the AI’s observations, not their own.

“When we compare notes, we tend to pull another person’s observations into our own memory,” Ferguson explained. The same could be said for the observations of an AI software.

“If the full narrative is whatever the body camera has picked up, that becomes the proposed truth of whatever happened. … At some point, why do we have officers testify at all?” Ferguson said. “It’s deeply offensive, in some sense, to think about a system in which humans are being prosecuted by cameras and whatever a software determines is relevant from what the camera picks up.”

This can be especially problematic if the transcription struggles to understand an accent or the meaning of slang terms. Further problems could emerge with AI’s video software. Ferguson explained that AI can have “difficulty with interpreting and understanding faces of minorities,” and that AI-attributed observations about a person’s actions, body language, behavior and facial expressions could be misunderstood — especially among minority communities. 

According to Axon, studies for racial bias in Draft One “could not detect a statistically significant difference in Completeness, Consistency or Word Choice Severity between races.” The studies evaluated whether AI-generated transcripts differed depending on race.

In April 2024, the company said a more detailed report about the studies would be released. However, The Park Record was unable to find additional information about the racial bias studies.

Lack of transparency was another of Ferguson’s concerns, since private corporations aren’t subject to the same transparency laws, like the Freedom of Information Act and the Privacy Act, as a government agency is. That could make it difficult for defense attorneys or the public to dig into the software to better understand how it works.

He was also skeptical that AI-drafted police reports would be properly reviewed by officers before they hit approval. 

“The problem isn’t an officer who’s able to condense an hour’s worth of work into 10 minutes. The problem is the officer who says, ‘I can do even better. I can condense this into 30 seconds.’ … You have to have your head in the sand to think that that’s not going to become normal among some officers, and even culturally accepted in some departments,” Ferguson said.

But Sever believes any faulty reports will be caught in the administrative process.

“Once (a police officer) signs off on (a police report), it goes to a supervisor. That supervisor is required to look at that report for completeness and accuracy, and then sign off on it,” he explained. “Then our records department gets the report, … and lots of times, they’re the ones that also kick back reports for something that they see.”

An officer who continually failed to review AI-drafted reports before submitting them would be disciplined and potentially terminated.

Looking forward

Sever said the only reason the department may not immediately implement one of these programs is if it doesn’t fit into the current budget cycle. 

“An officer in the station does not affect slowing people down on the road. They’re not available to respond to a call in a timely manner,” Sever said. “When we can get them out of the station, on the road, is when the public is going to see them, and that’s when it changes people’s behaviors for the better.”

Feedback from officers seems to have been positive as well.

“One of my sergeants, he’s not a big tech guy, but he’s super excited about this,” Sever said. “He goes, ‘I came back to the station. This report would have taken me an hour, and it took me eight minutes.’”

The post Heber City Police Department test-pilots AI software appeared first on Park Record.

You might have noticed the shift start about May last year. 

That’s when Google started implementing its AI Overview feature, which showed users an AI-powered, aggregated response to searches rather than just listing relevant websites. 

“What was going on is that you’re reading the derivative rather than the original,” said Stephanie Cohen, a Parkite and the chief strategic officer for Cloudflare, an information and communications technology company. “So we were going from a world of search engines to a world of ‘answer engines.’”

Suddenly, you didn’t have to click through to a website to get the information you were searching for, even for answers to relatively complex questions. AI Overview would come up with one, courtesy of all the websites it had indexed, free of charge to Google.

And before that, ChatGPT, the popular large language model — or AI, if you want to feel like you live in the future — first launched in November 2022. The service reached 100 million users by January 2023, according to The Guardian, and like many popular AI models, it can also return synthesized answers to user queries based on trawling the internet.

Whatever you think of that, Google and AI companies have for the last few years benefited massively, treating the internet like a free buffet by scraping information off websites and condensing it into pithy answers — before anyone could say otherwise.

The end result, which is probably here to stay, is a shift in how we use the internet, likely more convenient for the user. But what’s the cost to the people who feed the machine? How are writers, artists and journalists footing the bill of this new paradigm, and can AI companies be made to pay their fair share? 

Take a newspaper like The Park Record, for instance. Original reporting, while still valuable as print media, has also become fodder for the aggregate machine of the online world. That’s a change to the internet’s longtime business model, one which Google had originally helped facilitate: a search engine points to a website, and the website can sell ads or charge a subscription — or not. That was the deal. 

As that dynamic has changed, Cloudflare is one company seeking to rebalance content creators’ leverage in the AI era. 

The California-based company works with customers to build and run applications “so they’re fast and secure,” Cohen said, claiming that more than 20% of the internet is built with Cloudflare. Small companies, big companies, free customers and paying customers, a lot of people rely on Cloudflare, including The Park Record. After all, Cloudflare’s CEO and co-founder is Matthew Prince, co-owner of The Park Record with his wife, Tatiana. 

Cloudflare’s services are the kinds of things that will make sense to you if you’re computer savvy, and sound like nonsense otherwise: content delivery networks, web access filters, DDoS mitigation — stuff that’s important to the internet running smoothly. 

For talking about AI, the company’s bot management services are most important, said Cohen.

“Basically, (we) make sure good traffic gets through as fast and securely as possible,” she explained, “and we deflect bad traffic, traffic you don’t want on your site. We use that same network to protect employees … your data, your network on the internet.”

Good in, bad out. Around when Google started implementing AI Overview, Cloudflare started hearing from its customers. 

“The way that traffic was coming to (customers’ sites) was changing, and my guess is that’s true for The Park Record; it was true for publishers and content creators of all sizes,” Cohen said.

They dug into the data and found that what customers were saying was exactly right, Cohen said. AI was eating up traffic. 

“And if you thought about it and you thought about your own personal life, what was going on is that you’re reading the derivative rather than the original,” she said. “But you still need the original source to have a thriving internet, to have an internet that helps you answer these types of questions. And so we realized that this was going to increasingly be a problem.”

More specifically, Prince, on a panel at the recent Fortune Brainstorm Tech 2025 — a tech-focused conference that Fortune Live Media on Sept. 8-10 at the Montage Deer Valley — said that the AI Overview feature has made it “10 times harder to get traffic from Google than it was just a few years ago.” 

“That’s the good news,” Prince said at the panel. “With OpenAI, it’s 750 times harder. With Anthropic, it’s 36,000 times harder to get traffic.” 

So, a year ago, Cloudflare announced AI Audit, which let customers see when AI bots were scraping their site for data and let them choose whether to allow it. It was an opening move meant to let users pause the sudden and rampant practice of AI crawling. Since then, over a million internet domains opted to “block all,” the most aggressive option, Cohen said. 

Then, on July 1 this year, Cloudflare declared it Content Independence Day, introducing a couple more important changes. 

“The first thing we did was we changed the defaults on the Cloudflare network,” Cohen said. “And so what does that mean for all new domains and new customers? We block AI training. It is in the signup flow. … You can decide whether you want to leave that on, or you can turn it off.”

Essentially, an LLM that’s in its early development phase needs information in order to build itself into a coherent and useful tool. This move gave users the option to block AIs from using their website as part of that training phase. 

The second thing was to increase the granularity of control that website owners have regarding the bots on their site, giving them the ability to allow or block them depending on what they are there to do, Cohen said. For example, when an LLM answers a user’s question, it uses the information it already knows and also tries to search the internet for the most up-to-date info, too. This move could allow that sort of traffic if the user wants it. 

The third and most eye-catching change was the introduction of a beta program for what Cloudflare calls “Pay Per Crawl” — essentially, a website owner can choose to let a bot onto their site to scrape their data if the bot’s owner will pay their asking price to do so, done via what’s called a “402 request.”

Cohen said there’s been a high demand for the pay per crawl functionality since it launched. To give a sense of the scale:

“Every day, there are a billion 402 requests on the Cloudflare network,” Cohen said. 

Billion, with a B.  

The 402 requests can be personalized to include information for a crawler to bring back to its owner about how to contact the content creator so humans can negotiate a price. The whole thing can also be automatic, allowing users to set a price per crawl at the point of the request, or a bot may be authorized to pay for up to a certain amount. It’ll fulfill the 402 request then and there in that case. 

“The idea around all this stuff is that we believe that the site owners have the right to transparency, to know who is on their site,” Cohen said. “They have the right to know what they’re doing on the site, and then they have the right to decide whether or not they should be (there), and maybe they want to allow them for free or maybe they want to charge them.”

This ability for creators to have a say over who can access content, rather than a blanket assumption from AI companies that they can access anything they want, creates scarcity, Cohen said. That then creates a new kind of marketplace — one in which original work, such as local journalism, can recover some of the work’s monetary value online that AI and other factors have eroded.

“I think local news is really interesting because it’s something that people really care about right in a specific area,” Cohen said. “I do think that it is highly valuable to a bunch of these AI companies to have really good consumer experiences. And having really good consumer experiences means having access to accurate, high-quality, up-to-date information.” 

On that same Fortune Brainstorm Tech panel, Neil Vogel, CEO of People Inc., which says it is the largest digital and print publisher in America, said that being able to block AI traffic gives content creators and publishers the necessary leverage to give them a seat at the table with AI companies like OpenAI. 

“Some of the AI shops are good actors,” Vogel said. “OpenAI, we have a commercial deal with them. We’re very pleased with it. We do a lot of work with them.”

Bill Gross was on the panel representing his AI company, ProRata.ai, which employs a revenue sharing model with content creators on a pay-per-use basis and markets itself as a solution to concerns about AI traffic. Gross likened the way AI companies crawl content to the early days of YouTube. 

“If you look back at … when YouTube first started, people were stealing content left and right and posting it. There were take down notices every day from the studios,” he said. “And then finally, YouTube said, ‘How about if we share revenue with you?’ And the studio said, ‘Great, we’ll give you all this stuff.’ And it opened up a whole range of explosions of great content.” 

Google’s AI, in contrast, Vogel said, is a particularly bad actor because it uses one AI crawler for everything it does: Google’s AI is the same if it’s searching a site to index it in order to show it in a return of search results or if it’s just seeking an answer for an AI Overview result. If you cut off the crawler at all, you lose search traffic, too. Right now, it’s all or nothing with Google, even with Cloudflare’s ability to give users more granular control over AI bots. 

Prince, who is openly critical of Google, was bullish: 

“My prediction is that by this time next year, Google will be paying content creators for crawling their content and taking it and putting in AI models,” Prince said.

His hope for the future of the internet is swiss cheese: as AI companies catalogue what does exist to scrape, they’ll also uncover what doesn’t exist — the holes in the model that reveal areas of potential interest for users.  

“And so I think the business model of the future is actually to say to content creators, here are the holes in the cheese, and if you fill those holes, then we’ll actually compensate you and pay you for that,” Prince said.

An internet built on compensating creators who fill those demands would be more beneficial to humanity than a purely traffic-driven model of the internet, Prince said, in which whatever grabs your attention is the most rewarded. That’s the dynamic Google has encouraged and profited from, sometimes at the expense of a healthy information ecosystem, he said. 

“Why do we have a rise in populism? Why do we have a more divided society?” Prince said. “In part, it’s because the business model of media has gone down the direction of, how do you get as much traffic as possible? There is an opportunity going forward that we can actually get back to rewarding original content creation.”

Another panelist, Janice Min, CEO of Ankler Media, which publishes multiple entertainment industry-focused newsletters via Substack, said in contrast to Vogel’s desire to work with AI companies that she was more skeptical of striking deals.

“Our value is that we can create information that people will pay a lot of money for,” Min said. “I don’t see the benefit to us, and I’m willing to be dissuaded, but I don’t see the benefit to us partnering with any AI company right now.”

Publishing is always hard, and the check from tech companies always looks good, Min said, but her experience in media over the years has told her not to negotiate with the giants. 

“What happened to publishing 10 years ago … the Metas, the Googles, they just wanted to come, made it attractive, and then they rolled the carpet back up and took everything, including, primarily, the advertising,” Min said. “So I feel like I have to defend our audience. I have to defend our content.”

According to Cohen, the AI answer model is probably here to stay. Whether it’s tenable is a matter of making creators equal players with the companies who want their content. 

“I think a lot of consumers prefer that customer experience (of AI answers). And so we’ve just got to figure out an ecosystem and a model that works for that. Because that only works insofar as you have really great, high-quality content you have access to, and so there’s got to be an incentive for that high-quality content to get produced,” she said. “So we’ve just got to figure out something that works.”

As for what the shift to pay per crawl could mean for local journalism, according to Don Rogers, editor and co-publisher of The Park Record, it could set a positive precedent on the business side, maybe leading to a healthier industry.

“The big problem for local journalism isn’t the journalism, but the business with the major aggregators taking our content and using it to help undercut the funding sources for it,” Rogers said. “They do not reinvest in the journalism, and so here we are, with three quarters of the journalists gone and replaced by commentators, basically.”

And, practically speaking, forcing AI companies to work out a fair fee for use is better than trying to sue them or turning to government solutions, Rogers said, such as with The New York Times, who sued OpenAI for using the Times’ content for model training. 

“To me, it suggests a more elegant solution than countries like Australia and Canada trying to tax a Facebook or a Google for the content they, in essence, steal,” he said.

As for how this change could impact The Park Record specifically, Rogers said the paper is “a drop in the proverbial ocean.” 

“Honestly, I think we’re too small to move any dials,” he said. “I just really like the implications that should encompass all the social media companies and content aggregators ripping off content and starving actual journalism at the root,” he said.

The post For AI, the internet was a free buffet — until now appeared first on Park Record.