Khoury News

Khoury team helps Oakland government employees navigate AI productivity and safety

Between a lack of expertise, stretched-thin resources, and compliance concerns, City of Oakland employees have some trepidation about adopting AI. Their neigborhood computer science college is lending a hand.

May 28, 2026

by Benjamin Hosking

Share article

Close this dialog window

Companion read: Khoury students put on AI workshops for Oakland businesses and nonprofits

When it comes to using artificial intelligence models like ChatGPT, Claude, or Copilot, users bring a wide range of experience with prompt engineering. While the tools offer myriad ways to enhance workflows and deliver better results, many resource-constrained organizations lack the in-house expertise to train their staff.

In 2025, Khoury students and faculty at Northeastern’s Oakland campus held a series of workshops with local organizations, including City of Oakland government departments, that were designed to do just that: improve employees’ AI literacy and help develop AI policies.

Assistant Teaching Professor Rasika Bhalerao led a workshop for Oakland’s AI working group, where they learned that the group’s members would receive a free deployment of Microsoft Copilot. Oakland’s chief information officer asked Bhalerao and fellow faculty member Jessica Staddon if Northeastern could follow along to see how AI is being used and whether to expand the deployment citywide.

The project is supported by a Google Research Award, which focuses on resource-constrained organizations using artificial intelligence. The City of Oakland is a prime example.

“They’re not staffed to develop their own models or maintain them,” Staddon said. “The question is how organizations like that can safely use AI. The Copilot deployment crosses many departments, with staff from the fire department, animal services, childcare, schools, and more.”

Data collected from Oakland participants — including surveys, interviews, and eventually data logs — informs Staddon and Bhalerao’s research into how local government staff are using AI and what stumbling blocks they’re encountering. They led a team of graduate research assistants in this work, including computer science master’s students and first-time researchers Vinal Dsouza and Rebecca Earle.

“Before this experience, research was a very new area for me,” Dsouza said. “Once I started it was easy, we could talk freely to the professors, conversations about how we build this, what is the research question we are catering to. I’m getting better at finding a needle in a haystack.”

“I started interviewing several participants this semester, talking with people across different departments,” Earle added. “Afterwards, we discuss as a group, and I’ve found those conversations enriching and thought-provoking.”

The group recently presented their findings to City of Oakland officials. Recommendations included specialized training programs tailored to department-specific use cases and data sensitivity levels, clearer personally identifiable information guidelines co-developed with city employees, and a longitudinal follow-up survey to track how AI perceptions and usage are evolving. 

The group noticed participants’ prompt engineering skills evolving the more they were exposed to AI tools. Even in cases where a participant did not fully understand the concept of prompt engineering, they quickly grasped certain integral elements, such as “don’t ask the AI everything at once” or “break the prompt down into a couple of tasks.” The range of skills surveyed was wide, but most participants showed improvement with usage.

One recurring challenge was data privacy and safety in a government context.

“A public institution is subject to public records requests,” Staddon explained. “There were a lot of concerns about putting anything into an AI, as it could be requested. A lot of folks are concerned about transcribing meetings, which is super useful, but they are reticent to do so or are wary of areas like brainstorming and strategy development with AI.”

Dsouza points out that many people are uncertain what is right and what is wrong when it comes to government policy on AI and records, and this lack of clarity prevents them from taking advantage of AI.

“If people don’t know what to do, they won’t use AI,” Dsouza said. “They worry about having access to things they’re not supposed to, like if AI reads an email chain that becomes accessible to other people in the organization.”

Although AI poses privacy and policy concerns, it can also help protect the public from scams and fraud. Computer science master’s student Lucia Sun works with Staddon on a project about scam and fraud detection for resource-constrained organizations. The duo developed a prompting framework that incorporates Federal Trade Commission and Consumer Financial Protection Bureau definitions of fraud to improve classification accuracy. They analyzed how personal information included in narrative prompts can influence model predictions and classification outcomes.

“Scams influence a lot of people’s lives,” Sun said. “I think it’s very important to have a way for these organizations to distinguish scams and protect themselves.”

Currently, local governments like Oakland’s use semi-automated ways to distinguish and categorize scams, with trained officers supported by computer programs. Sun is working on a way to use large language models to help categorize and distinguish which scams need a deeper look and which will not, reducing the human personnel needed.

“I’m very interested in how AI spreads rumors and misinformation, and how AI-generated content influences the news and internet,” Sun added. “I’m working with [Staddon] on a new project about how to protect against an LLM disclosing redacted information like address and age.”

Through a separate practicum course led by Bhalerao, Oakland undergraduates also work directly with local partners, including teams from Oakland’s government, to implement AI.

“The most prominent one that comes to mind is the illegal dumping project,” Bhalerao said. “A group of undergraduates will create a computer vision neural network to detect illegal dumping that can be run on cameras already around the city.”

Although the challenges posed by AI to city governments like Oakland’s are numerous — from automated scams to unauthorized data disclosure — the potential for AI to help improve outcomes is palpable, provided there are clear usage and privacy policies in place.

“It’s a tough situation for strapped organizations like many cities,” Staddon said. “AI could really help in some ways.”

“There have to be clear ways they can be using it in their specific job,” Earle added. “Clarity and consistent guidelines are needed for everybody to be more confident in using it.”