The administrative machinery of the modern state was built for a slower, more deliberate world. Forms required human effort to fill. Complaints demanded time, patience, and a working knowledge of bureaucratic language. That world ended, quietly, sometime in late 2022. The arrival of large language models and the AI agents built on top of them has triggered an unprecedented, and largely unnoticed, surge in applications, petitions, and complaints flooding public institutions across the globe. This is not a future scenario. It is happening now.
The Data Behind the Flood: A Fivefold Surge in Complaints
The numbers tell a stark story of compounding change. In the United Kingdom, complaints to the Housing Ombudsman more than doubled between 2022 and the end of last year, jumping from 2,600 to over 7,000. Across the Atlantic, the United States’ Consumer Financial Protection Bureau (CFPB) recorded a fivefold increase in complaints over the same period. Similar explosive growth has been documented in Brazilian judicial petitions and German parliamentary petitions. These are not isolated anomalies; they are the leading edge of a structural shift in how citizens engage with the state.
Researcher Chris Schmitz has been systematically tracking this phenomenon. In a paper scheduled for presentation at the upcoming AI Ethics and Society conference, Schmitz documents 84 distinct cases of what he terms “agentic flooding” across 11 different jurisdictions. The dataset, hosted on his website for public reference, reveals a consistent pattern. Across welfare applications, official judicial appeals, and public consultations, submission volumes remained roughly flat before 2022. Then, as AI technology began to diffuse into mainstream usage, the numbers started to rise. Crucially, in most of the 84 cases, the growth has not slowed. The trajectory suggests this is not a spike but a permanent escalation.
What Is Agentic Flooding? The Mechanism Behind the Metrics
Agentic flooding describes a specific type of systemic overload. It is not primarily about malicious actors or spam bots, though those exist. It is about the drastic reduction in the “friction cost” of interacting with bureaucracy. Filling out a complex benefits form or drafting a formal complaint used to require significant cognitive effort. It involved reading instructions, gathering documents, composing text, and navigating labyrinthine online portals. For many people, that effort was prohibitive. They had a legitimate claim, but the administrative burden was simply too high.
AI agents, especially those powered by models like ChatGPT and Claude, have collapsed that friction cost to nearly zero. As Schmitz notes, the process has become incrementally easier. “People are finding out that this is something one can do,” he explained. “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.”
This ease of use is the engine of the flood. The technology does not need to be perfect. It only needs to be good enough to turn a task that was previously abandoned into a task that is quickly completed.
Real People, Legitimate Claims: The Critical Distinction
The most important nuance in this story is the quality of the surge. The instinctive reaction of many administrators has been to dismiss the new filings as spam, an AI-generated nuisance similar to the low-quality, AI-generated bug reports that flooded security bounty programs last year. Those reports, as documented by TechCrunch, were largely worthless, yet companies were still obligated to vet them, creating a significant drain on resources.
Public services could easily face a similar resource crisis, processing five times the volume of applications with the same budget and staff. But Schmitz’s research reveals a critical difference. “The vast majority of cases we find are people who are entitled to claim for something, claiming for that thing,” he told TechCrunch. The applicants are real. Their claims are legitimate. The underlying need was always there. The only thing that changed was the barrier to entry.
This distinction is everything. It transforms the problem from a cybersecurity nuisance into a profound policy challenge. The state is not being attacked by a flood of nonsense. It is being overwhelmed by a wave of previously unmet, valid demand.
The Role of Administrative Burden in Suppressing Demand
The concept of “administrative burden” has long been understood in policy circles. Research published in the Journal of Public Administration Research and Theory has shown that the costs of learning about, complying with, and navigating bureaucratic processes systematically deter eligible people from accessing services. These burdens are not neutral; they function as de facto gatekeeping mechanisms. They disproportionately affect those with less time, less education, or less access to professional advice.
AI is dismantling these barriers at a speed that public institutions are not designed to match. The result is a sudden, massive release of pent-up demand. The Housing Ombudsman in the UK, the CFPB in the US, and the judiciary in Brazil are now facing the full, unfiltered volume of grievances and applications that were always there but never formally submitted.
Why This Is Different From a Bot Attack
A bot attack is characterized by automation, repetition, and malice. The intent is often to overwhelm a system, to probe for weaknesses, or to generate noise that hides real threats. Agentic flooding is different. It is characterized by augmentation, not pure automation. A human being is still in the loop, but they are using an AI agent as an executive assistant to handle the arduous work of form-filling and text composition.
This subtle shift has enormous implications for how public institutions should respond. A simple rate limit or captcha system, designed to stop automated bots, is unlikely to be effective. It would block the legitimate user who is using an AI to write their complaint. It would also do nothing to address the underlying administrative structure that created the bottleneck in the first place.
The growth in volume mirrors the trajectory seen in other domains. Bug bounty programs were an early warning system. They showed that when you make a submission channel frictionless, the quantity of input skyrockets, even if the average quality per unit declines. The lesson for public services is that they are dealing with a structural change in user behavior, not a temporary glitch.
A Rare Opportunity to Remake Social Services
Schmitz’s research offers a perspective that is surprisingly optimistic. While the immediate challenge is daunting, he sees the current moment as a rare opportunity to fundamentally rethink how public services are designed. For decades, these services have been shaped by the assumption that the citizen bears the primary burden of navigating complexity. The form, the portal, the process—all of it was built with an implicit friction cost that acted as a throttle on demand.
AI has removed that throttle. The demand is now visible. The policy question is no longer “How do we process the same number of people more efficiently?” but rather “How do we redesign services for a world where everyone is an assisted applicant?”
“A big part of making AI go well is being able to detail out what the good version of things looks like,” Schmitz says. “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.” He believes this could be the moment for a complete rethink. “This could be the moment to say, ‘we need to rethink pretty much everything about how this process looks.’”
What a Redesigned Service Would Look Like
A service designed for the AI era would not simply accept AI-generated forms. It would be built from the ground up around an API-first architecture. Instead of presenting a human-readable web form that an AI must interpret and fill, the service would offer a direct, structured interface that a software agent could call natively. The human applicant would state their intent in natural language, the AI would translate that into the structured data the system requires, and the result would be validated and submitted automatically.
This would eliminate the current inefficiency of an AI scraping a web page, interpreting the fields, and then generating text that fits the boxes. It would also eliminate the need for human staff to spend their time parsing AI-generated prose that is often verbose and formulaic. The goal would be a state of “bureaucratic symbiosis,” where the human, the AI agent, and the government system work as a coordinated whole.
The technical capability to do this has existed for years. What has been missing is the institutional will and the recognition that the old model is no longer sustainable. The flood of applications is making that recognition unavoidable.
The Resource Crisis: Managing Five Times More Applicants With the Same Budget
The most immediate and painful consequence of agentic flooding is the resource strain. Public institutions operate on fixed budgets that are set years in advance. A fivefold increase in application volume cannot be absorbed without a corresponding increase in processing capacity. The result is longer wait times, staff burnout, and a growing backlog that undermines the legitimacy of the entire system.
This is not a problem that can be solved by hiring more people. The rate of growth is outpacing the ability of any government to recruit and train new staff. The only sustainable solution is to leverage the same technology that caused the problem to manage the solution. AI agents can be deployed on the processing side as well, triaging applications, checking for completeness, and flagging cases that require human judgment.
The comparison to the bug bounty crisis is instructive. Companies that were overwhelmed by AI-generated reports responded by adjusting their triage processes, often using their own AI systems to filter the incoming submissions. Public services will need to adopt a similar strategy. They will need to invest in AI-powered case management systems that can handle the volume, or they risk complete institutional paralysis.
Geographic Scope: A Global Phenomenon With Local Variations
The data from Schmitz’s research makes clear that this is not an American or British problem. It is a global phenomenon. The paper examines 84 cases across 11 jurisdictions, including Brazil, Germany, the UK, and the US. The pattern is remarkably consistent. In every jurisdiction with a digital service that can be accessed by an AI, the volume of submissions has increased since late 2022.
The specific services affected vary by country. In Brazil, the surge has been most visible in judicial petitions. In Germany, it is parliamentary petitions. In the UK, it is housing complaints. In the US, it is consumer financial protection complaints. The common thread is that all of these services had an online portal and a low barrier to entry. The AI did not need to hack the system; it only needed to use the system as intended, but at a scale and speed that was previously impossible.
This geographic diversity highlights the universality of administrative burden. It is a feature of all modern states, regardless of political system or culture. And AI is a universal solvent for it.
What Is the Correct Institutional Response?
The hardest question, and the one that Schmitz’s paper leaves open for policymakers, is what to do in response. There is no easy answer. The traditional tools of bureaucracy—forms, waiting periods, verification steps—are the very mechanisms that AI is designed to overcome. Adding more friction to the system would punish the legitimate applicants that the technology is now helping.
One approach is to explicitly recognize AI-assisted applications and create a separate, faster processing channel for them. This would accept the reality of the flood and treat it as an opportunity to increase service uptake. Another approach is to mandate that all applications must be submitted by a human acting without assistance, but this is practically unenforceable and would return the system to the pre-2022 status quo of suppressed demand.
The most viable path forward lies in redesign. Instead of trying to stop the flood, institutions should build the floodgates, channels, and reservoirs to manage it. This means investing in digital infrastructure that is designed for machine-to-machine communication. It means training staff to work alongside AI agents, rather than against them. And it means accepting that the era of the form is over. The future of public services is conversational, asynchronous, and deeply, fundamentally assisted.
The flood of AI-generated applications is not a bug in the system. It is a signal. It is telling us that the old methods of gatekeeping through friction are obsolete. The question is whether our public institutions have the imagination and the will to build something better in their place.
