States are embracing AI to help manage safety-net programs
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Sarah Grillo/Axios
States are increasingly deploying artificial intelligence to help manage their social safety net programs and reduce worker caseloads, but advocates warn there's little evidence of safeguards around the technology.
Why it matters: AI systems can hallucinate and make errors, and without proper oversight, a machine's mistake regarding program eligibility could strip Americans of essential benefits.
The big picture: President Trump's One Big, Beautiful Bill has helped push state government into the AI era, Allison Buffett, senior health policy analyst at the Bipartisan Policy Center, tells Axios.
- Some states are rolling out AI-assisted chatbots to answer Medicaid beneficiaries' eligibility questions, Florida lawmakers included an AI system to check a user's Supplemental Nutrition Assistance Program (SNAP) eligibility in its 2027 budget, and New Hampshire officials are working with Google Gemini to streamline how applicants submit information when starting an unemployment claim.
- Next year, programs like Medicaid and SNAP will adopt stricter work requirements and more frequent eligibility recertifications for applicants. Some of those time-consuming tasks are being given to AI rather than government workers, Buffett said.
Yes, but: A small discrepancy in data, such as slightly misreporting income on different aid applications, can boot people from benefit programs they are eligible for.
Case in point: TechTonic Justice Founder Kevin De Liban, who successfully sued Arkansas over its use of an algorithm to determine Medicaid home-based care benefits, says the system led to a drop in eligible hours for thousands of residents and operated in a "haphazard and harmful" way.
- While litigating the case, he found the state hadn't done projections on the magnitude of the cuts, and had "nobody on staff who could explain how the algorithm worked mechanically until over a year and a half into using it."
- De Liban says he ultimately figured out the algorithm on his own, and that it worked in "patently absurd ways that don't at all jive with what our regular understanding is of how to meet somebody's care needs."
Zoom in: Constant testing, monitoring and piloting AI programs in safe settings is critical, Code for America CEO Amanda Renteria tells Axios.
- She says trying out a program in small counties before scaling statewide is useful, as is watching for sharp eligibility drops, which could signal an AI system needs correction.
Oversight issues are further complicated because AI developers don't have to explain how their product works; many argue that the information is proprietary data.
- That can contribute to people who were wrongly denied benefits going months without compensation, even if they regain eligibility.
Zoom out: When an eligible person receives a wrongful determination saying they're ineligible for their benefits, that can be "devastating" to them and their families, Buffett says.
- "They will lose access to essential services that allow them to live in their communities, contribute to society, access health care, and live healthy and productive lives."
The bottom line: De Liban says he is "sympathetic" to governments seeking "increased efficiency," but adds that it is "unacceptable" for unchecked harms to fall on the nation's most "vulnerable."
- "State officials, no matter what circumstances they're finding themselves in, either need to vet these technologies appropriately, or they need to not use them and use a lower-tech way of administering their programs."
Go deeper: Revenge of the AI bubble
