Public services around the world are seeing a significant increase in applications and other requests, as AI makes it easier to fill out forms and submit complaints.
In the UK, complaints to the Housing Ombudsman have more than doubled since the introduction of ChatGPT, rising from 2,600 in 2022 to just over 7,000 last year. According to the U.S. Consumer Financial Protection Bureau (CFPB), the number of complaints increased fivefold over the same period. There were similar spikes in judicial petitions in Brazil and parliamentary petitions in Germany.
Researcher Chris Schmitz tracks this increase as part of a broader trend he calls the “agent flood.” In a paper to be presented at next month’s AI Ethics and Society conference, he examines 84 different cases of potential flooding across 11 jurisdictions and finds widespread evidence that AI tools are changing the way people interact with public services.
The more difficult question is what to do in response. While some new applications are clearly adversarial, others are the result of legitimate applicants using AI to file claims that would otherwise be abandoned. While some may view the new application as AI-generated spam, Schmitz sees it as a valuable opportunity to reimagine social services for the AI era.
Schmitz looked at services ranging from welfare applications to formal judicial appeals, all of which have online services that can be accessed by AI assistants. (For reference, the complete dataset is hosted here). For methodological reasons, Schmitz’s paper does not say that AI is directly causing the surge in new applicants. However, most of the 84 cases followed the same basic arc, with the number of filings remaining roughly flat until 2022, then increasing at an accelerating pace as AI technology becomes more widespread. Importantly, in most cases there has been no slowdown in growth, suggesting that growth is likely to continue in the coming years.
For Schmitz, it’s easy to see how AI usage will increase as AI skills and availability improve.
“People are starting to realize that this is something they can do, and gradually it’s getting easier. Before, you might have had to drag around a lot of context and prompt ChatGPT 3.5 very precisely, but now you might just paste a letter or take a photo in the Claude app and get a pretty good response in one shot,” he said.
This spike in volume is similar to what many bug bounty services experienced last year, as companies found their inboxes flooded with low-quality reports generated by LLMs. Although reports rarely contained serious security issues, companies were still required to scrutinize them, which was a significant drain on resources. It is easy to imagine that the public service faces similar problems as it manages five times as many applicants with the same budget.
But Schmitz said that while the bug bounty program was flooded with worthless submissions, most new applications for public service came from genuine people with legitimate claims.
“The majority of cases we find are people who have a right to claim something and are making a case for it,” he told TechCrunch.
If these people had not previously claimed benefits, it may be because the application process was too difficult. This is known in the policy world as administrative burden. With AI able to ease that burden, there could be an opportunity to rework many of these services in a more AI-friendly way.
It’s a big challenge, one we’ve barely begun to tackle, but Schmitz believes it’s a reason for optimism.
“The key to making AI work is being able to explain in detail what a good version of things looks like, and anyone who has used ChatGPT to prepare a tax return knows that there is a good version here that can help you,” Schmitz says. “Now may be the time to say, ‘We need to rethink almost everything about what this process looks like.'”
If you make a purchase through links in our articles, we may earn a small commission. This does not affect editorial independence.
