For decades, Kansas ran its unemployment insurance system on software that predated the internet. Agents kept five screens open at once, one of which was so old it could only be operated by typing commands, no mouse allowed. If you lost your job and tried to file a claim at nine at night, the system was often shut down for overnight batch processing, and you were out of luck until morning.
That changed at the end of 2024, when Kansas replaced the core system and started layering AI tools on top of it. An internal chatbot now searches the agency's own forms, policies, and records in real time while a staff member is on the phone with a claimant, instead of making them dig through files or put the caller on hold. The state says this dropped benefit processing times by nearly 80 percent, and more than 90 percent of claims are now resolved without ever needing a live agent.
This is not an isolated case. Michigan launched an AI chatbot for its unemployment agency this year, and Georgia has been running one long enough to report it correctly routes the vast majority of requests.
North Carolina and California have shipped similar tools in 2025 as well. State unemployment offices, some of the most outdated corners of American government, are quietly becoming one of the more active testing grounds for AI in the public sector.
There is a reason this wave is happening now and not five years ago. During the pandemic, states processed a record flood of claims on systems built in the 1970s, and the result was a well documented disaster.
Investigators later estimated that over 190 billion dollars in pandemic unemployment payments were improperly paid, a large share of it outright fraud, because outdated systems could not verify claims fast enough. Congress responded by sending states over a billion dollars in federal grants specifically to modernize this technology. Kansas's overhaul is one visible result of that money finally landing.
The fraud angle is actually the more consequential part of this story, even though the chatbot gets the headline. Kansas says the same AI tools now flag suspicious applications and scan for security weaknesses that hackers could exploit.
Given how much money was stolen from these exact systems last time, catching bad claims before they are paid out matters more to the state budget than shaving minutes off a phone call. For anyone who assumes government offices are always the last to adopt new technology, this is worth a second look.
A recent survey of state technology chiefs found the large majority now have formal AI policies in place, and adoption has moved well past experimentation. If you deal with any state agency, benefits, licensing, permits, tax offices, expect faster response times over the next two years, and expect that agency to be quietly using AI to check your paperwork before it ever reaches a person.