AI Tools Are Helping People Claim Benefits They Were Always Owed, and Public Agencies Aren't Ready

For a lot of people, the hardest part of getting help from the government has never been whether they qualify. It's filling out the paperwork. Anyone who has tried to appeal a denied insurance claim, file a housing complaint, or request a benefit knows the particular exhaustion of forms that ask the same question five different ways, deadlines buried in fine print, and phone trees that lead nowhere. That friction, it turns out, was quietly keeping a lot of eligible people from ever applying.

Now that friction is disappearing, and public systems are struggling to keep up.

In the United Kingdom, complaints to the housing ombudsman have more than doubled since ChatGPT's public debut, climbing from 2,600 in 2022 to just over 7,000 last year. The U.S. Consumer Financial Protection Bureau saw complaint volume grow fivefold over the same stretch. Brazilian courts and the German parliament have both reported similar spikes in petitions. These aren't isolated blips. They're part of a pattern researcher Chris Schmitz has taken to calling "agentic flooding," and he's built a dataset around it that spans 84 cases across 11 countries.

Schmitz's paper, set to be presented next month at the AI Ethics and Society conference, doesn't claim to prove AI caused every one of these surges. That's a hard thing to prove definitively, and Schmitz is careful about the limits of his methodology. But the pattern across almost all 84 cases is strikingly consistent: submissions stayed roughly flat before 2022, then began climbing as AI tools became more widely used, and that climb hasn't slowed down in most cases. If anything, it looks like it's still accelerating.

The mechanism behind this isn't mysterious. It's the same reason more people file their own taxes now than a decade ago, just applied to government forms instead of tax software. "People are finding out that this is something one can do, and incrementally, it is just getting easier to do it," Schmitz told TechCrunch. A few years ago, getting useful help from a chatbot meant carefully assembling context and writing a precise prompt into something like GPT-3.5. Now you can photograph a confusing letter from a housing authority, hand it to Claude, and get a workable response in one try.

The Difference Between Flooding and Spam

There's an obvious comparison here to something that happened in cybersecurity last year. Bug bounty programs, which pay researchers to find security flaws, got buried under a wave of low-quality reports generated by large language models. Companies were legally and practically obligated to review every submission, even the junk ones, and that review process became a serious drain on staff time and budget. It's not hard to picture public agencies facing the same fate: five times the applications, the same number of caseworkers, and no extra funding to bridge the gap.

But Schmitz's research points to something different happening in public services. Unlike the bug bounty flood, which was mostly noise, the new wave of applications appears to be substantively real. "The vast majority of cases we find are people who are entitled to claim for something, claiming for that thing," Schmitz said. These aren't fabricated claims or automated spam. They're legitimate requests that, in a lot of cases, would have gone unfiled without AI assistance.

That distinction matters enormously for how policymakers should respond. Policy researchers have a name for the barrier that used to stand between eligible people and the benefits they deserved: administrative burden. It's the compliance costs, the learning costs, the psychological toll of navigating a bureaucracy that wasn't designed with the applicant's convenience in mind. Administrative burden has always functioned as a kind of invisible rationing mechanism. Programs stayed within budget partly because a meaningful share of eligible people simply never applied, worn down by the process before they ever got an answer.

AI is now dismantling that rationing mechanism, and it's doing so faster than most agencies can adjust their staffing or their budgets. That's not a small technical hiccup. It's a structural shift in who actually accesses the safety net that governments have promised to provide.

Why It Matters

The instinct in a lot of institutions will be defensive: treat the surge as noise, add friction back into the system, maybe require in-person verification or make the forms harder for a bot to parse. That would be a mistake, and arguably a cruel one. If most of these new applicants are people rightfully claiming benefits they were always owed, then adding friction just reconstructs the barrier that AI helped tear down, punishing exactly the population these programs were meant to serve.

Schmitz sees a better path, though he's clear it's a heavy lift. "A big part of making AI go well is being able to detail out what the good version of things looks like," he said. "This could be the moment to say, 'we need to rethink pretty much everything about how this process looks.'" That means agencies designing intake systems that assume applicants will show up with AI assistance, building in verification steps that scale without requiring a caseworker to manually review every submission, and funding staff levels that match actual demand rather than historical filing rates depressed by red tape.

None of this is guaranteed to happen well. Governments move slowly, and budgets are political long before they're practical. But the alternative, letting agencies drown quietly while people who finally found their voice get buried in backlogs, would waste a genuine opportunity. The tools that helped someone draft a housing complaint or a benefits appeal are the same tools that could help agencies process them fairly and quickly, if the institutions are willing to rebuild around that reality instead of resisting it.