Artificial intelligence is making public systems easier to approach—and harder to operate at their previous scale. A study of 84 potential cases across 11 jurisdictions finds that applications, complaints and petitions have surged as AI tools have become widely available.
Researcher Chris Schmitz calls the pattern “agentic flooding.” His paper, scheduled for presentation at the AI Ethics and Society conference next month, does not claim that AI directly caused every increase. It instead documents a recurring shift: submission volumes were broadly flat before 2022, then began rising faster as AI spread.
Public Requests Rise Across Jurisdictions
Complaints to the United Kingdom’s housing ombudsman increased from 2,600 in 2022 to just over 7,000 last year. Over the same period, complaints to the US Consumer Financial Protection Bureau grew fivefold. The research also identifies similar increases in Brazilian judicial petitions and German parliamentary petitions.
The 84 cases range from welfare applications to formal judicial appeals. What connects them is an online process that an AI assistant can help a person navigate. Schmitz said a task that once required gathering context and carefully prompting an earlier chatbot may now be as simple as photographing a letter and requesting a response from a current assistant.
Most of the cases in the dataset have not shown a slowdown. Schmitz therefore expects volumes to keep rising, although that is a projection rather than a measured outcome. Because the study is observational, other changes affecting public-service demand may also contribute to the increases.
More Volume Does Not Necessarily Mean Spam
Public agencies face a familiar operational problem: every submission must be assessed, even when the budget and staffing remain unchanged. Bug-bounty programs experienced a comparable wave of low-quality reports generated with language models, forcing companies to investigate many claims that contained no meaningful security issue.
Schmitz argues that public-service filings are different. In the cases he studied, he says the vast majority came from people entitled to make the claim they submitted. That distinction reframes the surge. If AI helps eligible citizens complete processes they previously abandoned, the higher volume exposes unmet demand rather than simply producing synthetic noise.
Policy researchers describe the time, knowledge and paperwork required to obtain a service as administrative burden. AI can lower that burden by interpreting letters, organizing facts and drafting forms. The gain for applicants becomes pressure for the institutions receiving them, particularly when systems were designed around a stable number of human-prepared submissions.
The evidence does not imply that every AI-assisted filing is accurate or beneficial. Some activity is adversarial, and agencies still need to distinguish valid claims from duplication, error and deliberate abuse. The central problem is that the same technology can expand legitimate access and cheapen mass submission at once.
Services Need Capacity and Better Process Design
Blocking automated assistance would risk restoring the barriers that prevented eligible people from claiming benefits or challenging decisions. Accepting unlimited machine-generated material without changing review processes could overwhelm staff. The study points toward a third response: redesign services around the reality that applicants increasingly arrive with AI assistance.
Schmitz views the moment as an opportunity to specify what helpful use should look like. That could mean rethinking how institutions collect information and assess claims, although the source does not provide a finished policy model, cost estimate or tested implementation. The work of adapting these systems has largely not begun.
Any redesign will need to preserve access while managing authenticity, repetition and workload. Agencies may also require clearer ways to identify the underlying human claim when an assistant has generated polished or excessive text. Those are practical implications of the documented trend, not conclusions tested by the paper.
Agentic flooding is therefore not one simple category of abuse. It describes a capacity mismatch created when the cost of asking falls much faster than the cost of answering. The study’s most important finding is not merely that numbers rose, but that institutions may be seeing requests that friction once concealed. Responding well will require public services to treat accessibility and operational resilience as the same design challenge.
