AI Agents Are Swamping Public Offices With More Claims — and Forcing a Reset

Updated September 10, 2026 6:24 pm

AI tools are making it easier for people to submit forms, appeals and complaints to government agencies, and public services are already feeling the strain. New research says the number of requests has jumped sharply across several countries since ChatGPT arrived, with the UK housing ombudsman, the U.S. Consumer Financial Protection Bureau and multiple courts seeing major increases.

The trend matters because it is not just about spam. Researchers say AI is also helping legitimate applicants overcome paperwork barriers that once kept them from claiming benefits, filing petitions or challenging decisions.

What is happening to public services?

Public agencies are being hit by a flood of digital requests that appears to be closely tied to the rise of consumer AI. The increase is showing up in complaints, petitions, appeals and benefit claims, and in many cases the volume has climbed far faster than the capacity of the institutions receiving them.

In the United Kingdom, complaints handled by the housing ombudsman rose from roughly 2,600 in 2022 to a little more than 7,000 last year. In the United States, complaints submitted to the Consumer Financial Protection Bureau increased fivefold over the same period. Researchers have also identified major jumps in Brazilian judicial petitions and German parliamentary petitions.

Those numbers are part of a broader pattern that researcher Chris Schmitz describes as “agentic flooding” — a term for the surge in submissions generated or facilitated by AI systems that can draft, structure and send requests at scale.

Why are AI tools changing public administration so quickly?

AI is lowering the effort needed to complete bureaucratic tasks that used to require time, confidence and familiarity with official procedures. A person who once needed to gather documents, write formally, and understand the structure of a complaint can now ask an AI assistant to help generate a first draft, summarize a letter or fill out a form from a photo.

That shift is especially important in services that are already hard to navigate. Many public systems rely on online portals or standardized forms, which are easy for software to interact with. Once an AI model can read instructions, prepare text and guide a user through a sequence of fields, a process that once felt inaccessible can become much easier to complete.

Schmitz said the change is being driven by a steady improvement in both the capability and convenience of AI tools, noting that tasks which once required extensive prompting and context gathering can now often be handled in a single pass from a document photo or simple upload.

The result is a system that can suddenly process much more demand without any corresponding expansion in staffing, budgets or technical capacity.

How big is the increase?

The best answer is that the rise is widespread, fast and still continuing. Schmitz examined 84 possible cases of flooding across 11 jurisdictions and found that many followed the same trajectory: volumes stayed relatively flat before 2022, then began climbing quickly as AI tools spread into everyday use.

Just as important, many of the cases have not yet shown signs of leveling off. That suggests the pressure on public agencies may intensify rather than fade in the near term.

Schmitz’s paper, which is scheduled to be presented next month at the AI Ethics and Society conference, focuses on a wide range of public-sector systems, from welfare applications to formal legal appeals. The full dataset is being shared publicly for reference.

How do researchers know AI is involved?

They do not claim to prove a simple one-to-one cause-and-effect relationship. Instead, the paper says the evidence is strong enough to show a broad association: submissions were stable before 2022, and then they began rising alongside the spread of AI systems that can generate text, summarize documents and guide users through forms.

That distinction matters. The study does not argue that AI alone created the surge. It argues that the timing, scale and consistency of the pattern across multiple systems make AI a plausible and increasingly important driver.

Schmitz’s analysis also points to a practical feature of modern public services: many already live online. If a portal can be accessed by a human, it can often be accessed by an AI assistant as well, which makes government services especially exposed to automated or semi-automated submissions.

What does “agentic flooding” mean?

“Agentic flooding” refers to a wave of requests made possible by AI agents or AI-assisted workflows that can act on behalf of users. In this context, the flooding is not necessarily malicious. It can include everything from fully automated submissions to people using AI as a helper for tasks they would otherwise struggle to finish.

The term draws a parallel with earlier waves of low-friction digital abuse, but the public-sector version is more complicated. Unlike obvious spam, some of the new filings are legitimate and deserve attention. That makes the problem harder to sort out, because agencies cannot simply dismiss the influx as junk.

Not all of the new requests are the same

Schmitz’s research suggests the traffic falls into two broad categories:

• Adversarial submissions, where AI may be used to overwhelm systems or create administrative noise.
• Legitimate claims, where AI helps real applicants file requests they were already entitled to make.

That split is central to the policy challenge. A government office can reject spam, but it cannot easily ignore a real benefit claim or a valid complaint just because an AI helped produce it.

Why this is different from bug-bounty spam

The pattern may look familiar to companies that run bug-bounty programs, which were inundated last year with large numbers of low-quality security reports written with the help of large language models. Those organizations still had to review each submission, creating real operational costs even when the reports were mostly useless.

Public services face a similar resource problem, but with a harder twist. In government, many of the requests are not garbage. They may be proper applications for housing help, consumer relief or legal review. That means the agency still has to triage, investigate and respond.

The burden is especially acute when funding and staffing stay flat while incoming applications multiply. A fivefold increase in complaints can quickly become a service-delivery crisis.

Why researchers see opportunity, not just risk

Schmitz argues that the surge should not be viewed only as a threat. In his view, it is also evidence that AI can reduce the “administrative burden” that keeps people from engaging with the services they are entitled to use.

Administrative burden is the policy term for the cost of paperwork, uncertainty, confusion and effort that discourages participation. For many citizens, the most difficult part of asking for help is not eligibility; it is the process itself. AI can remove enough friction to turn a failed attempt into a completed application.

Schmitz has said that the key challenge is to define what a successful AI-assisted process should look like, and that the current moment could force governments to rethink the design of many routine procedures.

That view turns the debate on its head. Instead of asking how to stop AI from touching public services, policymakers may need to ask how to redesign those services so they remain fair, accessible and manageable when AI is part of the workflow.

How should governments respond?

The short answer is that there is no single fix. Governments will likely need a mix of technical, procedural and policy responses.

Possible responses include stronger identity checks, better form validation, smarter triage systems, clearer application guidance and more automation on the agency side. But each of those tools has drawbacks. Stricter verification can reduce abuse, yet it can also make services harder to access for the people they are meant to help.

That trade-off is the core tension in the story: AI can expand access and create overload at the same time.

1. Reduce friction for legitimate users. Simplify forms and language so fewer applicants need AI just to complete basic steps.
2. Improve intake systems. Use automated screening to separate routine cases from urgent or high-risk ones.
3. Protect against abuse. Add rate limits, anomaly detection and verification for unusually high-volume submissions.
4. Redesign services for an AI world. Build public portals that assume many users will arrive with machine assistance.

What happens next?

The pressure on public services is likely to keep building as AI systems become more capable and easier to use. Schmitz’s research suggests the current surge is not a temporary spike but an early sign of a structural shift in how people interact with institutions.

That means the next phase is not only about managing volume. It is about deciding whether government workflows should continue to assume a human at every step, or whether they need to be rebuilt for a world where many applicants arrive with an AI helper in hand.

For policymakers, the lesson is uncomfortable but clear: AI is no longer only changing how people write emails, summarize notes or search the web. It is also reshaping who shows up at the front door of the state, how often they arrive and how difficult it is to process them.

Timeline of the emerging pattern

What began as a convenience feature for drafting text is now becoming a governance problem. The scale of the shift is still emerging, but the direction is already visible: AI is making it easier to ask the state for something, and the state is struggling to catch up.

Whether that produces a better, more accessible public sector or a wave of costly administrative overload will depend on how quickly agencies adapt.

For now, the evidence suggests the flood is real, the trend is growing, and governments have only just started to respond.

Source note

This report is based on research by Chris Schmitz and reporting on public-service request volumes in the United Kingdom, the United States, Brazil and Germany.