Imagine a world where the tools we use to talk to each other become so good at writing, reading, and organizing that they can handle the paperwork of daily life for us. This is the promise of artificial intelligence agents: software that can help a person understand a complex rule, fill out a form, or write a letter to an official. For decades, experts have known that when it is easier to ask for something, more people will ask for it. If a government makes a benefit application simpler to find and fill out, more eligible people will apply. This is a basic principle of how people interact with public services. But now, a new question has emerged. What happens when the tools that make these interactions easier become so powerful that they can generate thousands of requests in a single day? The concern is not just that more people will apply, but that the sheer volume or complexity of these requests could overwhelm the government offices designed to handle them, potentially breaking the system or forcing officials to make the process harder for everyone.

A team of researchers set out to investigate whether this scenario is already happening. They coined a term for the phenomenon: "agentic flooding." This describes a sudden surge in the number or complexity of requests to government services that is caused by artificial intelligence agents and that strains the capacity of those services. To understand the scope of the problem, the researchers did not rely on speculation or computer models. Instead, they went looking for real-world evidence. They scanned government services across eleven different countries, including the United States, Germany, the United Kingdom, and Australia. They looked for any instance where officials or credible news sources had pointed to a spike in demand and explicitly linked it to the use of artificial intelligence.

The result was a collection of eighty-four specific cases where this link was made. The researchers found that this phenomenon is already occurring across a wide range of government domains. The most common pattern they observed was surprisingly simple. In the vast majority of cases, the surge was not caused by fully autonomous robots navigating websites on their own. Instead, it was driven by large language models—advanced text-generating tools—creating sophisticated legal or administrative text for humans to submit. For example, in one case in Germany, social courts saw a massive rise in lawsuits where the letters submitted were thousands of pages long, generated by AI but filed by people. In another instance in Australia, officials noted a wave of requests for information that appeared to be written by machines. The data suggests that the primary mechanism is humans using AI to write better, longer, or more frequent applications, which then enter the system and take up the time of human workers.

The researchers then asked which services are most at risk. By analyzing their dataset, they identified a clear pattern. The services most vulnerable to flooding are those that offer a high financial reward but have historically been difficult to access because the application process is complex or confusing. When AI agents lower the cost of navigating that complexity, the hidden demand is unlocked. Tax returns, court claims, and applications for disability benefits fit this description perfectly. These are services where a successful application can mean a significant payout, but where the paperwork is often so daunting that many people give up. The study suggests that as AI makes the paperwork easier, the number of applications for these specific services is likely to spike, putting pressure on the systems that process them.

To help governments prepare, the researchers developed a way to assess the risk for any given service. They created a framework that looks at two main things: how easy it is for an AI to interact with the service, and how hard it is for the government to handle the result. If a service accepts long, free-form text and has few limits on how many times a person can apply, it is highly susceptible. If the government office processing the request has a rigid budget, a slow manual process, or a legal requirement to read every single word of every application, the impact of a surge will be severe. The study indicates that while the current strain is manageable, the risk is highest for these financially attractive, complex services.

The paper also explores how governments might respond to such surges. Historically, when demand for a service spikes, officials often try to slow it down by adding "friction." This could mean charging a fee to apply, requiring an in-person visit, or limiting how many requests a person can send. The researchers found that these measures are often effective at stopping the surge quickly. However, they come with a heavy cost. Adding friction makes the service harder to use for everyone, not just the AI users. It can disproportionately hurt poorer people or those who are less comfortable with technology, creating an unfair barrier to essential services. In some cases, it can even undermine the right to access justice.

The authors argue that while these quick fixes work, they are not the best long-term solution. A better approach is to build the capacity of the government to handle the demand. This could involve redesigning the service to be more efficient, using digital identity to verify who is applying, or even using AI on the government side to help process the routine parts of the work. The study suggests that governments have a choice: they can react to a surge by making life harder for their citizens, or they can proactively redesign their systems to handle the new reality. The researchers recommend three immediate steps. First, governments should audit their services to see which ones are most vulnerable. Second, they should integrate digital identity systems into the most exposed services to make it easier to verify applicants and limit abuse. Third, they should clarify the laws around these measures so they know what they are allowed to do before a crisis hits.

The study concludes that while the current situation is not yet a total collapse of government services, the trend is clear. The era of AI agents making government interaction easier is here, and with it comes the risk of being overwhelmed. The research does not predict a future where machines take over the government, but it does warn that the current way many services are designed is fragile. If governments wait until the surges become unmanageable, they will likely be forced to choose between letting the system break or making it harder for people to get help. The paper suggests that by acting now to understand the risks and redesign the systems, governments can avoid this difficult trade-off and ensure that the benefits of artificial intelligence are shared by all, rather than becoming a burden that excludes the most vulnerable.