Pittsburgh Governs AI Use by City Workers Without Public Policy
The city has spent three years building internal AI guidelines and training staff, but hasn't released the rules publicly even as police and other departments deploy the technology.
Pittsburgh has been quietly governing how its employees use artificial intelligence for three years, but the public still doesn't know what those rules say.
City workers began using generative AI tools like ChatGPT as soon as they became available in 2023, according to Andrew Hayhurst, Pittsburgh's senior manager of innovation. Rather than block the technology, the city chose to create usage standards. An early restrictive policy became public in 2024 through reporting, but the updated 2025 version remains internal.
Today, the city designates Microsoft 365 Copilot Chat as its recommended tool because data entered stays within Pittsburgh's Microsoft system and isn't used to train AI models. Employees receive guidance on which tools to use, what information to exclude, and when to disclose AI assistance.
"The city's focus has been familiarizing city staff with the guidelines," rather than making them publicly available, city press secretary Molly Onufer said.
Why it matters
As AI tools become embedded in government operations—from drafting documents to analyzing data—the lack of public transparency creates accountability gaps. This matters most in high-stakes areas like policing, where Pittsburgh has already violated its own policies by using facial recognition software after Black Lives Matter protests. Without published guidelines, residents can't know how AI affects decisions about public safety, services, or their personal data.
The transparency problem in policing
Pittsburgh Public Safety didn't respond to questions about what AI tools the police bureau currently uses or whether formal policies govern them. According to Beth Schwanke, executive director of the University of Pittsburgh's Institute for Cyber Law, Policy and Security, common law enforcement AI applications include license plate readers, gunshot detectors, and automated police reports.
The city does use some AI-powered tools. Pittsburgh Public Safety has deployed Traffic Jam, a platform from local company Marinus Analytics, since 2015. The tool uses computer vision to scan online sex ads and images for trafficking investigations. Cara Jones, the company's cofounder, said police typically cite evidence the tool finds without naming it.
Jones argues some opacity protects investigations, but Schwanke warns that without governance structures, even effective AI systems will erode public trust.
Building policy through employee input
The city hired Tia Christopher, founder of AI consulting firm the Orange Peel Collaborative, to educate employees on current standards and potentially develop them further. Christopher started with a sentiment survey asking what city workers think of the AI rollout, what training they need, and what tasks they want to automate. After more than 200 responses, she's conducting listening sessions and plans to release an internal case study this month.
"[An AI] policy—for, really, private sector, city, anything—it starts with listening and asking questions," Christopher said.
The approach reflects a three-year process involving research, stakeholder meetings, and a cross-department generative AI working group, according to Christopher.
The speed problem
AI's rapid evolution distinguishes it from past technological shifts. "Five years in AI is an eternity," said Vincent Conitzer, a Carnegie Mellon University computer science professor and AI ethics expert. "Five years ago I would not have believed where we are already today."
That speed creates challenges for governance. Schwanke expects the novelty to fade within five years, giving governments clearer insight into which uses need new rules, which fit existing oversight, and when AI shouldn't be used at all.
Meanwhile, bias and privacy remain leading ethical concerns. AI systems trained on historical data can amplify existing inequalities in lending, policing, or hiring. "Not only is it reflecting our existing biases back, it can also make them worse because there is this feedback loop," Schwanke said.
Countering these risks requires a unified public citywide policy, vendor transparency about how tools work, and enough flexibility for different agency needs, according to Schwanke.
These details were first reported by Alice Crow for PublicSource as part of the Pittsburgh Media Partnership's Pittsburgh250 collaborative project.
This is an original analysis by the Omega editorial team. Source reporting: AI Watch.
Want systems like this working for your business?
Book a Call