AI-assisted filings are surging across public services, raising pressure on agencies
A new study finds AI-assisted filings are surging across public services, easing administrative burdens for claimants while straining agencies and budgets.
AI tools are helping more people complete forms, submit complaints, and pursue official appeals, producing sharp increases in requests to public services in several countries. The growth is giving eligible claimants access to systems they may previously have avoided, but it is also forcing agencies to process far more submissions without evidence that staffing and budgets are keeping pace.
Researcher Chris Schmitz is documenting the trend in a paper scheduled for presentation at the AI Ethics and Society conference next month. His review covers 84 potential cases across 11 jurisdictions, including welfare applications, judicial petitions, parliamentary petitions, and consumer complaints. He describes the pattern as “agentic flooding,” although the research stops short of proving that AI directly caused every increase.
The findings matter beyond government administration. They offer an early view of what happens when AI assistants reduce the time, knowledge, and confidence required to navigate complex institutions—and when those institutions are not designed to absorb a sudden increase in demand.
Complaint volumes are rising across jurisdictions
The examples identified by Schmitz show unusually large increases after 2022, as consumer-facing AI tools became easier to use. In the United Kingdom, complaints to the housing ombudsman rose from about 2,600 in 2022 to just over 7,000 the following year, according to the TechCrunch report. The United States’ Consumer Financial Protection Bureau recorded five times as many complaints over the same period, the report said.
Similar growth appeared in Brazilian judicial petitions and German parliamentary petitions. Across the broader dataset, submissions were generally stable before 2022 and then began increasing more quickly as AI adoption spread. Most of the cases had not yet shown a clear slowdown, suggesting that demand could continue to build.
Those figures should be treated as evidence of correlation rather than a definitive measurement of AI’s causal effect. The paper’s methodology does not establish that AI was responsible for each new submission, and the available reporting does not provide a case-by-case breakdown of how many filings were created or assisted by an AI system.
The term “AI agents” also covers a wide range of behavior in this discussion. Some people may use ChatGPT or Claude to draft a response, interpret a letter, or identify a relevant form. That is different from an autonomous system independently navigating a government portal and submitting a claim. The source evidence primarily describes AI-assisted activity, not a uniform population of fully autonomous agents.
Lowering administrative burden can reveal unmet demand
Schmitz told TechCrunch that most of the cases he reviewed involved people who were likely entitled to claim something and were pursuing that legitimate entitlement. That distinction is central to the policy implications. A larger queue is not necessarily evidence of abuse if AI is helping people overcome the administrative burden that previously stopped them from applying.
Government forms and appeals often require applicants to collect documents, understand specialized language, present a coherent account, and meet procedural requirements. For someone with limited time, legal knowledge, or confidence, those steps can be enough to abandon a valid claim. An AI assistant can reduce that friction by turning a letter or photograph into a draft explanation, suggesting missing information, or helping an applicant understand the next step.
The result is a difficult trade-off. Public agencies may face more work because more people can access them, but that increase may represent improved access rather than low-quality demand. Treating every AI-assisted filing as spam could deny service to the very applicants that digital tools are helping for the first time.
At the same time, the source compares the situation with the wave of low-quality, large-volume reports that some bug-bounty programs received after the spread of large language models. Public services cannot assume that every submission is meaningful simply because it comes from a real person. They still need ways to identify duplicates, incomplete applications, malicious activity, and claims that require human review.
Agencies face a capacity and reliability problem
The immediate operational risk is volume. If agencies receive several times more requests while retaining the same budget and staffing, backlogs can grow, response times can lengthen, and frontline workers may spend more time triaging than resolving cases. A system built around manual review may become less reliable precisely as more residents learn how to use it.
AI could help agencies manage that pressure, but deploying it safely is not a simple reversal of the applicant-side problem. Automated triage would need to avoid disadvantaging people whose documents are incomplete, whose language differs from the system’s training data, or whose cases do not fit common patterns. Public bodies would also need clear rules for privacy, recordkeeping, appeals, and human accountability.
For product teams and civic-technology builders, the evidence points toward workflow redesign rather than a thin chatbot layer. Useful systems would need to explain eligibility, request only necessary information, identify uncertainty, preserve supporting evidence, and make it easy for a human official to review the reasoning. Applicants should be able to correct an AI-generated draft before it becomes part of an official record.
The evidence supports opportunity, not a simple success story
The strongest claims in this story come from Schmitz’s ongoing research as described by TechCrunch, not from an official cross-government dataset or an independently verified causal study. The paper’s full dataset is reportedly available, but the supplied evidence does not establish its sampling method, the precise role of AI in each case, or whether the increases were driven by changes in policy, public awareness, staffing, or online access.
The reported complaint figures are also presented as examples of a wider pattern rather than as a controlled experiment. They show that public-service demand changed during the same period that generative AI became more accessible. They do not, on their own, prove that AI-generated submissions account for the increase.
That uncertainty does not make the trend unimportant. Even a partial contribution from AI would be consequential for agencies that have historically depended on administrative friction to limit demand. The key question is whether institutions can distinguish genuine access from abusive volume without rebuilding the barriers that AI has removed.
What to watch next
The next important signal will be the release and discussion of Schmitz’s paper at the AI Ethics and Society conference. Researchers and agencies will need more granular evidence linking individual submissions to AI assistance, while separating drafting help from autonomous action.
Public-service operators should also disclose changes in backlog, processing time, repeat submissions, and rejection rates as AI-assisted demand grows. Those measures could show whether the new volume reflects valid unmet need, procedural confusion, or automated duplication.
For builders, the practical indicators will be whether agencies publish rules for AI-assisted applications, offer machine-readable forms, and provide supported paths for human review. Clearer policies around identity, consent, document handling, and appeal rights will determine whether AI reduces friction without weakening due process.
Creati.ai perspective
The reported surge is a warning that AI adoption does not stop at the interface. When a tool makes an existing process easier, it can change the number and type of people who enter that process. Public agencies therefore need to plan for demand expansion, not merely automate their current queues.
The most constructive response is neither to reject AI-assisted filings nor to accept every submission automatically. Agencies should redesign services around transparent assistance, strong validation, and accountable human decisions. If the majority of new requests are legitimate, the challenge is not suppressing access—it is building public systems capable of handling the access that AI makes possible.
