A study identified 84 cases across 11 jurisdictions where cheap AI-generated text was linked to surges in filings to government services, a pattern the authors call “agentic flooding.”. The paper said complaints to the UK’s housing ombudsman rose from 2,600 in 2022 to just over 7,000 in 2025 after ChatGPT’s release, while the U.S.
Consumer Financial Protection Bureau recorded five times as many complaints over the same period.. Petitions to the Brazilian judiciary and the German parliament also increased sharply. The paper was co-authored by Chris Schmitz, Lewis Hammond of the Cooperative AI Foundation and Alan Chan of GovAI, and is due to be presented at the AAAI Conference on AI, Ethics, and Society on October 12 to 14..
The researchers said the growth has taken two forms: more filings and longer filings. Of the 84 cases, 50 showed quantitative flooding, 76 showed qualitative flooding, and 42 showed both.. One submission to a German social court exceeded 4,000 pages.
The paper said daily caps on submissions can limit high volumes, but do little when a single applicant files one very large document.. The study said the same underlying capability explained 87% of the surges in its sample: language models can generate text cheaply and at scale. The authors said no AI agents currently exist that can autonomously navigate agency websites..
The researchers started with 2,288 government services across 12 countries and kept only 84 cases that met three conditions. Each case required a plausible mechanism by which AI reduced the cost of applying, evidence that demand shifted, and an explicit statement from officials or another credible third party that AI caused the increase..
Judicial and legal services accounted for 19 cases, followed by regulatory complaints with 10 and welfare and social security with 9. In 58 of the 84 cases, a government official explicitly attributed the increase in filings to AI.. The paper said services with payouts or benefits tied to successful claims, including tax returns, small claims and property-value appeals, faced the highest near-term risk based on a 13-factor matrix.
It also said the findings do not prove causation across government services as a whole and should be treated as proof that the phenomenon exists, not as a full measurement of its scale.. The authors said governments could respond by adding fees, rate limits or in-person requirements, or by expanding capacity with more staff, their own AI systems or redesigned intake processes.
In 14 of the 84 cases, agencies responded by reinstating fees or blocking IP ranges, and Australia has considered reintroducing fees for Freedom of Information requests.. The paper warned that such measures can disproportionately affect poorer and less digitally literate applicants.
Separately, the Bank of England has warned that autonomous AI systems could threaten financial stability.. Featured image credit. Tags: Agentic flooding
