AI Agents Are Flooding Public Services With New Requests, Study Finds

AI tools are making it easier for people to file complaints, applications, petitions, and appeals, creating a new challenge for public services designed around much higher administrative barriers.

Researcher Chris Schmitz describes the trend as “agentic flooding.” His research examines 84 potential cases across 11 jurisdictions, including welfare applications, judicial filings, consumer complaints and parliamentary petitions that can be submitted through online systems.

The paper does not conclude that AI directly caused the increases. Instead, Schmitz found a recurring pattern:h submission volumes remained relatively stable before 2022 and then began rising as generative AI tools became more widely available.

Public agencies are seeing sharp increases in requests

In the United Kingdom, complaints to the housing ombudsman increased from about 2,600 in 2022 to more than 7,000 last year. Complaints submitted to the U.S. Consumer Financial Protection Bureau grew roughly fivefold over the same period, while similar increases appeared in Brazilian judicial petitions and German parliamentary petitions.

Schmitz has published the underlying collection of potential agentic flooding cases, allowing the broader pattern to be examined across different government services and jurisdictions.

AI assistants can reduce the work involved in understanding official correspondence, assembling information and drafting responses. A task that once required significant time and expertise can increasingly begin with something as simple as uploading or photographing a government letter and asking an AI system for help.

That creates a difficult distinction for public agencies. Some automated submissions may be low-quality or abusive. Still, Schmitz said many of the new filings appear to come from real people pursuing benefits, appeals or other claims they are entitled to make.

“The vast majority of cases we find are people who are entitled to claim for something, claiming for that thing,” Schmitz said.

AI may be reducing administrative burden

Public policy researchers have long studied administrative burden, including the learning, compliance and psychological costs people face when interacting with government programs. Complicated forms, unfamiliar procedures and the effort required to challenge decisions can discourage eligible people from completing legitimate claims.

Generative AI can lower some of those barriers by helping users interpret documents and prepare submissions. That can improve access to public services, but it can also sharply increase workloads for agencies operating with staffing and budgets built for lower volumes.

Schmitz compares the problem with the surge of AI-generated submissions experienced by bug-bounty programs. Security teams have reported spending time reviewing large numbers of low-quality reports because each submission still requires evaluation.

Public services face a different problem because many of their additional submissions may be legitimate. Simply blocking or discouraging AI-assisted filings could therefore prevent people from accessing services they previously couldn’t navigate.

Government systems may need to adapt

The research suggests that the challenge is not only how governments identify automated submissions, but how public systems should operate when AI makes complex administrative processes easier for ordinary people to use.

If submission volumes continue rising, agencies may need to reconsider workflows that depended partly on procedural difficulty to limit demand. Schmitz argues that AI could instead provide an opportunity to redesign services around the assumption that people will increasingly use assistants to understand rules, prepare forms and challenge decisions.

That leaves governments with a growing operational problem but also a chance to make public services more accessible. The central question is how agencies can handle substantially greater demand without treating legitimate AI-assisted applications as spam.