A study published Aug. 17 identified 84 potential cases across 11 jurisdictions in which artificial intelligence may be increasing the volume or complexity of requests reaching government services. The emerging pattern, which received new attention in a Sept. 10 TechCrunch report, has consequences well beyond government: AI is lowering the cost of asking for service faster than many organizations can lower the cost of delivering it.

The researchers call the phenomenon “agentic flooding.” Their examples include benefits applications, consumer complaints, court filings, information requests and public comments. In many cases, a person still submits the request; a language model simply makes it easier to understand the process and produce the required text.

That distinction matters. A larger queue is not necessarily an attack, and an AI-assisted request is not necessarily invalid. It may represent demand that existed all along but was suppressed by confusing forms, specialist language and the time required to navigate a bureaucracy.

The evidence shows pressure, not simple causation

The paper by Chris Schmitz, Lewis Hammond and Alan Chan is a preprint scheduled for presentation at the AAAI/ACM Conference on AI, Ethics and Society. It offers a structured dataset and a risk framework, not proof that AI caused every increase the researchers documented. The authors describe the cases as potential flooding, and TechCrunch noted that their method stops short of establishing a direct causal link.

Official data nevertheless show the scale of the operational change. The U.S. Consumer Financial Protection Bureau reported receiving about 6.6 million complaints in 2025, up from roughly 1.7 million in 2023. About 5.98 million of the 2025 complaints were sent to companies for review and response. The bureau’s own database cautions that complaint volume alone does not measure the prevalence of consumer harm.

The numbers therefore support a capacity problem, not a single explanation. Awareness, market conditions, organized campaigns, easier digital access and AI-assisted drafting can all affect volume. Leaders should resist using an AI label as a shortcut for deciding whether a request is legitimate.

Administrative friction was doing hidden work

Many service systems have been sized around the number of people who finish a difficult process, not the number entitled to use it. Long instructions, unstructured evidence requirements and intimidating language acted as unofficial filters. Generative AI weakens those filters by translating notices, organizing facts and drafting appeals or complaints.

That is beneficial when it helps a qualified person claim a benefit or challenge an error. It becomes costly when the receiving organization must review longer, repetitive or poorly grounded submissions. The same tool can expand access and create rework at the same time.

The study warns that quick responses such as fees, identity checks or narrower channels can reduce demand but may also exclude the people a service is meant to help. Rate limits are also a blunt instrument when the problem is not machine-speed submission but a growing number of people sending individually plausible, AI-assisted requests.

The business version is already visible

Customer support, insurance claims, warranty requests, vendor questionnaires, procurement bids and compliance appeals share the same operating logic. When AI makes a complex interaction cheap, more customers and counterparties will complete it. Historical abandonment rates stop being a reliable planning assumption.

For business leaders, this changes demand forecasting. A company cannot treat every increase as abuse, but it also cannot allow unlimited generated text to consume expert review time. The correct response is to make intake more structured while preserving a path for unusual or high-stakes cases.

Forms should request specific facts, evidence and desired outcomes instead of rewarding volume. Systems should detect duplicate claims, verify identity where necessary and route work by risk and complexity. Queue dashboards should measure not only arrivals, but also handling time, missing evidence, repeat contacts, escalation rates and the percentage of submissions that produce a valid action.

Capacity becomes part of the AI strategy

Organizations deploying AI on the customer side of a process should examine the receiving side at the same time. A chatbot that helps thousands of people draft requests can create a downstream obligation for operations, legal, compliance and customer-care teams. Improving the front door without redesigning the back office merely moves the bottleneck.

AI can help triage, summarize and compare submissions, but automated rejection creates a different risk. High-impact decisions need audit trails, review paths and clear accountability for errors. The goal is not to restore the old friction. It is to build enough capacity and structure that easier access does not collapse the service.

Agentic flooding is best understood as an operating-model warning. AI turns effort into volume. Leaders who once relied on paperwork, confusion or customer fatigue to regulate demand now need explicit rules, better interfaces and measurable service capacity. The queue is becoming a strategic system, and it must be designed for a world in which asking is nearly free.