AI Lowers Barriers Helping People Claim Benefits
Something unexpected is happening in the back offices of public services across the world. The forms are being filled. The complaints are being filed. The petitions are being submitted. And the people doing it are not who you might expect.
For decades, policymakers have quietly acknowledged a problem they rarely named in public: the administrative burden. The paperwork. The deadlines. The confusing eligibility rules. The phone lines that close at 4:30. These were not neutral features of a well-functioning bureaucracy. They were filters — and they filtered out the people who needed help most.
The Numbers That Tell a Different Story
In the United Kingdom, complaints to the Housing Ombudsman more than doubled after the arrival of accessible AI tools, rising from 2,600 in 2022 to just over 7,000 last year, according to the Housing Ombudsman’s published case data. The United States Consumer Financial Protection Bureau saw complaints grow fivefold over the same period, per its Consumer Complaint Database. Brazilian courts recorded similar surges in judicial petitions, and German parliamentary petitions followed the same curve.
The easy reading is that this is spam. A flood of machine-generated noise. But the researcher who has tracked these cases most closely, Chris Schmitz, reached a different conclusion. In a paper presented at the AI Ethics and Society conference, he examined 84 cases of what he calls “agentic flooding” across 11 jurisdictions His finding was not that bots were gaming the system. It was that real people were finally getting through.
“The vast majority of cases we find are people who are entitled to claim for something, claiming for that thing,” Schmitz told TechCrunch
That single sentence reframes the entire debate. What looks like a crisis of volume is, in many cases, a correction of a decades-old failure.
The Burden Nobody Wanted to Name
The concept of administrative burden has been studied for years in policy circles. It has three components: learning costs, compliance costs, and psychological costs. Learning costs mean figuring out whether you qualify. Compliance costs mean gathering documents, filling forms, meeting deadlines. Psychological costs mean the shame, the stress, the feeling of being judged.
For a single mother working two jobs, the learning cost alone can be prohibitive. For an elderly person with limited digital literacy, the compliance cost is a wall. For someone who has been rejected before, the psychological cost is enough to stop them from trying again.
These costs were never accidental. In some cases, they were designed to ration access. In others, they simply accumulated like sediment — layer after layer of well-intentioned rules that together became impenetrable.
The result was a silent injustice. People who were legally entitled to housing support, financial redress, or welfare benefits never claimed them. Not because they did not need them. Because the process was too forbidding.
What Changed
The change did not come from a government reform. It came from the tools people already had on their phones.
Schmitz describes the shift in practical terms. A few years ago, using an AI assistant to draft a complaint required dragging together context, crafting precise prompts, and hoping the model understood. Today, it can be as simple as taking a photo of a letter and pasting it into an app. The response comes back coherent, structured, and ready to send.
“It is just getting easier to do it,” Schmitz said. “Before it might have been a question of a lot of dragging context together and prompting ChatGPT 3.5 very precisely, it may now be a question of just pasting or taking a photo of a letter with your Claude app and getting a pretty good response in one shot.”
That is the concrete gain. Not a revolution. Not a utopia. Just a lower wall.
The Bug Bounty Parallel That Proves the Point
Critics point to what happened with bug bounty programs as a warning. When large language models became widely available, companies that paid for security vulnerability reports found their inboxes flooded with low-quality submissions. The reports were often worthless. But companies were obligated to review each one, draining resources.
The parallel seems obvious. Public services, after all, face the same math: more applications, same budget.
But the parallel breaks down at the crucial point. In bug bounty programs, the flood was mostly noise. In public services, Schmitz found the opposite. The applications were coming from people with legitimate claims — people who had been entitled to support all along but had been locked out by the complexity of asking.
A bug bounty program filters for a rare skill: finding security flaws. A welfare application filters for need. These are not the same thing. When the barrier to entry falls, the people who flow through are not opportunists. They are the intended beneficiaries.
The Opportunity Hiding in the Flood
Schmitz sees something more than a problem to be managed. He sees a rare opening.
“A big part of making AI go well is being able to detail out what the good version of things looks like,” he said. “And anyone who’s ever used ChatGPT to do the tax return knows that there’s a good version here where you’re being helped.”
That good version is not hypothetical. It is already visible in the data. People are using AI to understand eligibility rules they never knew existed. They are using it to draft appeals they would never have written. They are using it to navigate systems that were designed, whether intentionally or not, to keep them out.
The question is not whether this is happening. The question is what public services do about it.
The Last Open Variable
The surge in applications will not slow down on its own. Schmitz’s data shows that in most of the 84 cases he studied, the growth curve has not flattened. It is still rising. As AI tools become more capable and more accessible, the volume will continue to climb.
This leaves public services with a choice. They can treat the flood as a threat and build higher walls — stricter verification, longer processing times, more friction. Or they can treat it as a signal that the old walls were too high in the first place.
The first option is easier. It requires no new thinking. It protects budgets and preserves the status quo. It also means that the people who are finally finding their voice will be pushed back into silence.
The second option is harder. It means redesigning intake systems for a world where AI assistants are the first point of contact. It means rethinking what verification looks like when the volume is five times higher. It means accepting that the old administrative burden was not a feature to be preserved but a failure to be corrected.
Schmitz frames it as a moment of decision. “This could be the moment to say, ‘we need to rethink pretty much everything about how this process looks.’”
That is the open variable. Not the technology. Not the volume. The willingness of institutions to recognize that the surge is not a malfunction. It is the system finally reaching the people it was built to serve.
The AI did not create the need. It removed the barrier. Whether institutions treat that as a threat or a correction will decide who gets heard next.
Sources
2. AI Ethics and Society conference
3. TechCrunch