Frontline Justice and Josef Partner on Multi-State Rollout of AI to Protect Access to SNAP Benefits

The federal Supplemental Nutrition Assistance Program, or SNAP, is a critical benefit that helps to alleviate food insecurity for vulnerable populations, supporting over 40 million low-income individuals, nearly three quarters of whom live at or below the federal poverty level.

But recent changes to the program enacted as part of H.R. 1 (aka the Big Beautiful Bill) have altered requirements, increased complexity, and led to a surge in wrongful denials. The law introduced new work requirements and a maze of exemptions, and Nikole Nelson, CEO of Frontline Justice, says the added administrative burden all but guarantees mistakes.

“Mistakes will be made,” Nelson said in an interview yesterday. “Lots of people are going to fall through the cracks because of these additional administrative burdens.”

Now, Frontline Justice, a national nonprofit dedicated to closing the access-to-justice gap through a new category of legal helpers called community justice workers, and Josef, a legal tech and automation company, are announcing the multi-state expansion of an AI assistant called Frontline Q.

The two organizations built Frontline Q to help justice workers more effectively assist families in navigating this complex system and challenging benefit denials. After a pilot of the tool in Alaska – where the community justice worker concept grew out of a project of the Alaska Legal Services Corporation – the organizations are now expanding it to justice workers in Arizona and Texas as well, with plans eventually to roll it out nationwide.

Answers On the Spot

Even before Frontline Justice brought AI into the picture, Alaska’s community justice worker program was drawing measurable results. Research conducted by the American Bar Foundation found that Alaska’s justice workers, most of whom volunteer their time through legal aid and other community organizations, had helped draw down $23 million in SNAP benefits for families in rural and remote parts of the state that otherwise would have gone undistributed.

That track record is what led Frontline Justice to partner with Josef last year to pair its justice worker training with an AI assistant. Frontline Q is not a consumer-facing tool. Rather, in each state, the AI assistant is paired with community justice workers and designed to scale their advocacy by providing instant, explainable answers to complex SNAP eligibility rules.

“It is not client facing, it is justice worker facing,” Nelson said. “It’s human in the loop. It’s supervisor moderated.”

Before Frontline Q, a justice worker fielding a question from a client often had no choice but to email a supervisor and wait – sometimes up to 20 minutes for a simple answer, Nelson said, and sometimes days.

With Frontline Q, that same question can be answered on the spot, during the same phone call or office visit.

“It shortened the response time,” Nelson said. “They could have an immediate answer that they knew was correct and they could help the client right then who was in front of them rather than having to wait.”

Sam Flynn, COO and co-founder of Josef, pointed to a justice worker in a remote community who discovered a “magic” phone number that reliably connected clients to a helpful, live person at the state benefits office.

Under the old system, that kind of institutional knowledge tended to stay with whoever found it. With Frontline Q, she could submit it, have it validated, and within days every justice worker in Alaska had access to it.

In another instance, a justice worker added an infographic she used to help clients determine whether their household fell above or below the federal poverty line. Once approved, it became available throughout the Frontline Q network.

“It’s kind of this beautiful, controlled, but also crowdsourced knowledge resource for justice workers nationwide,” Flynn said.

Built on Josef, Tailored by State

Frontline Q is built on Josef, a legal automation platform used by major law firms and corporations, as well as by legal services and access-to-justice organizations. Its Josef Q, a product developed in partnership with OpenAI, lets legal teams turn policies and regulations into self-service Q&A tools.

(I wrote last year about the use of Josef’s technology to create Roxanne, a first-of-its-kind AI-powered tool created to help tenants navigate issues related to housing repairs.)

Flynn said the fit between Josef’s underlying technology and Frontline Justice’s needs became apparent almost as soon as he and Nelson met last year.

“There was this beautiful synergy between the knowledge architecture that informs Josef Q and the knowledge architecture required here,” he said.

SNAP is a federal program, but each state administers it under its own rules, and that variation shapes how Frontline Q is built for each new state.

“There is some localization that needs to happen, but we can work from a universal backbone and then move from there,” Nelson said.

To expand into a new state, Frontline Justice first localizes its training curriculum – typically a lightweight, on-demand course that takes justice workers two to three hours to complete – and then partners with an existing legal aid organization that already runs a SNAP or public benefits practice and is building out its own community justice worker program.

In Arizona and Texas, Frontline Justice is working with legal aid providers that are doing exactly that. Each state will also partner with a designated local organization to provide standardized training, onboarding, and supervision for their justice workers, and users can provide direct feedback to the system, allowing their insights to continually improve the model.

A Concept that is Gaining Ground

The rollout comes as the community justice worker concept is gaining ground nationally. Thirteen states plus the District of Columbia now have a regulatory environment that allows community justice workers, Nelson said.

These justice workers are trusted local advocates such as health workers, librarians and social service staff trained to provide legal help where traditional legal aid is unavailable.

Momentum is building to expand the justice worker concept nationally, Nelson said. Montana and the District of Columbia authorized justice worker programs within a month of each other, she noted, and Virginia, Maryland, New York, Wisconsin and California are among the states now considering similar policies.

“I really think that over the course of the next year we’ll have — I hope within the span of the next three years — all of the U.S. and the Virgin Islands and Puerto Rico will all have justice worker implementing policies in place,” Nelson said. “I don’t think that’s pie in the sky, given the rapid pace of momentum and how far we’ve come.”

Alaska’s program remains the largest in the country, with more than 200 community justice workers and growing. Nelson said the existence of a state-authorized justice worker program is not a strict prerequisite for deploying Frontline Q, but it reflects Frontline Justice’s mission of expanding who can deliver help, rather than training lawyers to do the same work.

Frontline Justice is funding the current rollout through its own grant-supported budget. The tool is being deployed through legal aid organizations at no charge, and Nelson said there is currently no fee-for-service model, though sustainability is something the organization is thinking about as it scales.

What Comes Next

Frontline Q currently addresses only SNAP, not Medicaid, even though the two programs face similar pressures under H.R. 1. Nelson said the decision to focus on SNAP was a matter of timing, because Medicaid’s new federal work requirements do not take effect until January. That gives Frontline Justice a runway to build out Medicaid-specific training before pairing it with Frontline Q.

Nelson said that the stakes of getting this right are illustrated by recent research showing that denials of health insurance coverage, both private and public, have risen 25% over the past decade. When people appeal those denials, they are overturned roughly half the time, but only about 10% of people appeal.

In Alaska, wrongful SNAP terminations spiked to an 87% error rate in 2023, before H.R. 1 took effect, leaving some families waiting up to six months for applications to be processed.

Both Nelson and Flynn described the SNAP and Medicaid rollout as a first step toward a broader vision. Nelson said she eventually wants to train people well outside the traditional legal aid system — food bank staff, librarians, church volunteers — so that someone who shows up needing help with a benefits denial can get it wherever they already happen to be.

Flynn said that as Frontline Q expands into more states, there could be another benefit beyond simply helping individual justice workers answer questions more quickly. The aggregated data could eventually offer a clearer national picture of where benefit systems are failing people.

Flynn also drew a distinction between Frontline Q’s supervised, source-verified model and the general-purpose AI tools people increasingly turn to on their own.

“These are not governed, these are not vetted resources,” he said. “People are very frequently getting incorrect responses to their questions, and that can lead to real risks.”

He pointed to vetted self-help projects such as Roxanne in New York City and efforts underway at Legal Aid North Carolina as examples of a different, more careful approach to self-serve legal AI — one Frontline Justice has deliberately stayed away from with Frontline Q.

“By embedding Josef’s technology into our justice worker model and rolling out Frontline Q into these three critical states, we are establishing a blueprint to scale access to justice nationwide,” Nelson said. “We are connecting trusted advocates with communities, ensuring that they are giving high quality legal guidance, and that families at risk of losing access to food get help faster than ever before. No one should lose access to food because they didn’t do the paperwork right.”