Pittsburgh250 — This story is part of a Pittsburgh Media Partnership collaborative reporting project exploring how Southwestern Pennsylvania communities are marking America’s 250th anniversary — and how local history continues to shape civic life today. Technical.ly, City Cast Pittsburgh and Pittsburgh’s Public Source teamed up to explore the region’s economic horizons.
By the time generative AI tools like ChatGPT went mainstream in 2023, Pittsburgh city employees weren’t waiting for permission to use them, according to Andrew Hayhurst, the city’s senior manager of innovation.
“Immediately, we start seeing people on our network using it,” Hayhurst said to a room full of technologists gathered in Bakery Square for a local innovation summit in April. “There was a big kind of scare and push and decision that had to be made. Do we block AI, or do we try to figure out a better way?”
The city decided to work with it, creating standards for generative AI usage. The early policy, which became public in 2024 through a series of Public Source articles, was largely restrictive, designed to curb risky behavior like putting private resident data into commercial AI tools.
Pittsburgh updated its policy in 2025, designating Microsoft 365 Copilot Chat as its recommended tool for city employees, meaning data entered into Copilot stays within the city’s Microsoft system and isn’t used to train AI models, according to the policy.
City employees are also being told which AI tools to use, what information to keep out of them and when to disclose that they’ve used the tech.
What are the main ethical concerns around city use of AI? Do Pittsburgh police use AI tools? Has AI use by police caused problems before? What is Pittsburgh’s current policy on AI use by city employees? Has the policy changed since then? Is the policy available to the public? What’s next for Pittsburgh’s AI policy?FAQ: Pittsburgh’s AI governance
Privacy and bias are the leading concerns, according to experts at Pitt’s Institute for Cyber Law, Policy and Security and CMU. Sensitive information can end up in systems not designed to protect it, and AI systems trained on historical data can carry and amplify existing inequalities in areas like lending, policing or hiring.
Pittsburgh Public Safety did not respond to questions about what AI, autonomous or algorithmic tools the Pittsburgh Bureau of Police currently uses, or whether those tools are governed by a formal policy. But we know from prior reporting that they use license plate readers and gunshot detectors.
Yes. When Pittsburgh police used facial recognition software during Black Lives Matter protests, the practice was found to have violated both bureau policy and city law.
The city created standards for generative AI use after employees began using tools like ChatGPT on their own after it went mainstream in 2023, according to the city’s senior manager of innovation. The original 2024 policy was largely restrictive, aimed at preventing risky behavior such as entering private resident data into commercial AI tools.
Yes. Pittsburgh updated its policy in 2025 to designate Microsoft 365 Copilot Chat as the recommended tool for city employees. Data entered into Copilot stays within the city’s Microsoft system and isn’t used to train AI models, according to the policy. Employees are told which AI tools to use, what information to keep out of them and when they must disclose AI use.
Not yet. The city’s focus has been familiarizing staff with the guidelines rather than making them publicly available, according to the city’s press secretary.
A local AI consulting firm is working with the city on what they describe as an ongoing three-year process to educate employees on current standards and possibly expand them. It has resulted in a cross-department Gen AI working group. An internal case study is expected this month on ways city employees could augment their work with AI.
Today, as officials set those guardrails, they’re also leaving room for experimentation, asking how AI could improve city work.
“[An AI] policy — for, really, private sector, city, anything — it starts with listening and asking questions,” Tia Christopher, founder of the local AI consulting firm, the Orange Peel Collaborative, told Technical.ly.
Christopher’s currently working with the city to educate employees on the current standards and possibly further develop them. She’s started with a sentiment survey, asking city employees what they think of the AI rollout, what training could be useful, what tasks they’re looking to automate and what outcomes they’re seeking.
“I don’t want to believe in that [future]. I would rather believe in smart tech, where we can all be more creative and all ‘people’ more.”
Tia Christopher, founder of the local AI consulting firm
After more than 200 responses, she’s moving to listening sessions, with plans to release a case study internally this month about how the city can augment but not replace employees with this tech. The alternative could be dystopian, she said, a future Pittsburgh that doesn’t learn from its labor movement past and treats humans more like machines.
“I don’t want to believe in that [future],” Christopher said. “I would rather believe in smart tech, where we can all be more creative and all ‘people’ more.”
MORE PITTSBURGH250 STORIES
Data centers could brown the landscape, or green the power sector
So, the city has moved from emergency restrictions to preferred tools to employee training in just a few years — with the aim to govern a technology that is already being used inside city operations — but the timeline for letting the public know about these evolving rules is still uncertain.
“The city’s focus has been familiarizing city staff with the guidelines,” rather than making them publicly available, Molly Onufer, city press secretary, told Technical.ly.
AI governance faces a transparency gap
Comparisons to past technological changes can be helpful, but only up to a point.
AI is moving much faster, according to Beth Schwanke, executive director of the University of Pittsburgh’s Institute for Cyber Law, Policy and Security.
With that speed comes more opportunities to erode public trust.
“No matter how good the AI is, or how many great systems we have in place,” Schwanke said, “if we don’t have the accompanying governance structures, I do not think people will trust government.”
The transparency gap is clearest in one of the highest-stakes areas of city government: policing.
Pittsburgh Public Safety did not respond to questions about what AI, autonomous or algorithmic tools the Pittsburgh Bureau of Police currently uses or whether those tools are governed by a formal policy.
“No matter how good the AI is, or how many great systems we have in place… if we don’t have the accompanying governance structures, I do not think people will trust government.”
Beth Schwanke, executive director of the University of Pittsburgh’s Institute for Cyber Law
License plate readers, gunshot detectors and automated police reports are among the most common AI applications in law enforcement, according to Schwanke, and some of those tools are already at work in Pittsburgh.
Pittsburgh police used facial recognition software soon after the Black Lives Matter protests — a practice later found to have violated both bureau policy and city law.
AI tools in policing aren’t new, according to Cara Jones, cofounder of the Pittsburgh-based tech company Marinus Analytics, but they’ve largely gone unnoticed until now.
Her company’s platform, Traffic Jam, uses computer vision — a form of AI — to scan online sex ads and images, helping investigators gather digital evidence for trafficking cases. Jones said the technology has been used by Pittsburgh Public Safety since 2015 and now operates across four continents.
Traffic Jam has supported local trafficking investigations, though police usually cite the evidence it finds rather than naming the tool itself, Jones said.
For Jones, that makes transparency necessary, but not absolute.
“We do need a certain level of transparency to help promote public trust — that we’re here to protect and serve for the greater defense and support of victims,” Jones said, “but if we overshare, there is a real risk that exploitation — and a criminal exploitation — learns how to sidestep those protections, and then we have to go back to square one.”
Bias, privacy and the limits of safeguards
Bias and privacy are two of the leading ethical concerns presented by the use of AI in government services, according to Schwanke and Vincent Conitzer, a Carnegie Mellon University computer science professor and AI ethics expert.
If AI tools are used carelessly or improperly, sensitive information can slip into systems that weren’t designed to protect it, but that’s just an amplification of the concerns governments already have when using digital tools, Conitzer said.
Bias can be harder to root out.
Even when an AI system is trained on data that looks representative of the real world, that data can carry the fingerprints of historical inequalities, such as discrimination in lending, policing or hiring, according to Schwanke.
“I kind of think of it as like a fun house mirror,” Schwanke said. “Not only is it reflecting our existing biases back, it can also make them worse because there is this feedback loop.”
Along with the city’s existing AI guidelines, some internal safeguards may already exist.
Jones from Traffic Jam said vendor vetting for AI-enabled tools across law enforcement agencies has become more formal over the last decade, including data privacy reviews and extensive questionnaires.
Auditing AI systems only works, however, if governments clearly define fairness and success up front, according to Conitzer.
Agencies also need to make the systems transparent enough to inspect and put someone on the hook for monitoring whether they’re doing what they’re supposed to do, Conitzer added. Many of those pieces are still a work in progress for local governments.
“We’re currently in a situation [where] most people are aware of AI to some extent, but detailed expertise is still pretty scarce,” Conitzer said. “We, as a society, also have to kind of catch up and figure out what we think is responsible behavior,” he added.
So, countering risks today requires a unified, public citywide AI policy, vendor transparency about how tools work and enough flexibility to account for agencies’ different needs, Schwanke said.
From what Christopher — the AI consultant — is seeing, that’s what’s happening at city hall.
“This has been a three-year process. They observed and did research and had appropriate meetings and stakeholder engagement. They have an ongoing Gen AI working group, where that’s cross-department,” she said, later adding, “[The policy] should be modular, and personally I think the more specific the better.”
Novelty will fade, but uncertainty remains
The rapid evolution of AI tools in public life has left municipal governments, as well as Fortune 500 companies, “really struggling to actually implement and figure out where AI is useful,” Schwanke said.
In the next five years, the novelty of these tools will likely wear off, she added, and as that happens, she expects governments to get a clearer sense of which uses call for new rules, which can fit into existing oversight systems and when AI tools shouldn’t be used at all.
Everything is moving so fast that it’s impossible to accurately predict 10 to 100 to 250 years from now, Conitzer said, but since detailed expertise on AI is still scarce, more oversight can’t hurt.
“Five years in AI is an eternity,” he said. “Five years ago I would not have believed where we are already today.”
This story was fact-checked by Emma Folts.
