The AI complaint flood: What record caseloads at courts and agencies really prove

What this is about

Whether it's an MP's office in London, a labour court in Berlin or an insurance ombudsman: everywhere, more complaints, objections and lawsuits are arriving than ever before, and the chatbot is often the prime suspect. Researchers have a name for it: “agentic flooding”. It describes a flood of submissions that happens because AI assistants and AI agents (programs that complete tasks on their own, such as filling in and submitting a form) make writing to the authorities almost free.

Over the weekend, the Guardian gathered voices from British public bodies under the headline “AI clog”. Hollie Wickens, a caseworker in a Labour MP's office, told the paper:

“Half of the emails we get are AI in some form.”

We put the German and British figures side by side and asked: which increase can actually be traced back to AI?

The numbers compared

Two examples show how easy it is to get the math wrong. For the Bundestag, a 34% rise sounds like a flood. But according to the Bundestag, 2024 was 18.8% below 2023. By our calculation, that means around 11,400 petitions arrived in 2023, and 2025 was only about 9% higher. At the UK Housing Ombudsman, the start of the increase in November 2022 coincides with more than just ChatGPT's launch. That same month saw the inquest into the death of toddler Awaab Ishak, who died from mould in a social housing flat. The ombudsman itself attributed the 2022/23 rise to the inquest, media coverage, poor housing conditions and changes in the law.

What's proven and what isn't

The paper “Characterizing Agentic Flooding of Government Services” by Chris Schmitz (Hertie School), Lewis Hammond (Cooperative AI Foundation) and Alan Chan (GovAI) is the most thorough survey so far. It collects 84 possible cases across 11 jurisdictions, including German social court lawsuits with filings allegedly running to more than 4,000 pages. The authors themselves write that their method allows no causal or quantitative claims. Only 25 cases have any time series at all, some rose before ChatGPT existed, and the paper only captures cases where someone explicitly blames AI. Speaking to beck-aktuell, Schmitz also called causality “extremely hard to establish”.

Our take: when it comes to volume, AI's share is documented almost nowhere. For the labour courts, social courts and travel arbitration, there are explanations that could largely account for the rise without AI. For the social courts, there's another factor the statistics don't show: since the “security package” at the end of 2024, asylum seekers in Dublin cases have been excluded from benefits. Numerous urgent applications were filed against this, and the federal government counted at least 80 urgent rulings against the exclusion. We don't know how much of the 47% rise this accounts for.

What is well documented, though, is the effort per case. Bodies report this independently of one another: - According to Deborah Evans of Lawyers in Local Government, complaints are on average about ten times as long as before. The longest ran to around 200 pages. - At the UK Financial Ombudsman Service (FOS), up to a third of responses to initial assessments in a small sample appeared to be AI-generated. In one case, 37 of 39 references were wrong, and some replies to a six-page decision ran to more than 200 pages. - The police watchdog IOPC reports complaints citing laws that don't exist in England and Wales. - According to ORF, Austria's data protection authority recognizes AI complaints “mainly by their length” and by targeted GDPR references. They “tie up considerable resources”.

One detail deserves special attention. The travel and transport arbitration body says many consumers only find it in the first place through AI. In that case, AI isn't just a typewriter but a signpost to rights many people didn't know they had. That's a burden, but not necessarily abuse.

Better justice for citizens?

The optimistic reading has the hardest number so far. A study by Minkyu Shin, Jin Kim and Jiwoong Shin analyzed more than 1.1 million complaints filed with the US Consumer Financial Protection Bureau (CFPB). Likely AI-assisted complaints led to relief in 49.3% of cases, compared with 39.9% for human-written ones. Using a statistical method designed to filter out chance effects, the authors estimate the effect at around 10 percentage points.

IOPC director general Rachel Watson doesn't see only downsides either: perhaps some people “didn't feel able” to complain before. The FOS stresses that AI particularly helps vulnerable and neurodivergent people put their case into words. Journalist Sam Freedman sums up the dilemma: complaining gets easier for each individual, but backlogs grow for everyone.

The flip side is errors. Consumer champion Helen Dewdney (“The Complaining Cow”) had chatbots write complaints. One pointed a broadband complaint to the regulator Ofcom instead of the relevant ombudsman scheme, while another named an ombudsman that doesn't exist. The IOPC is also seeing a falling success rate: in 2022/23, 39.5% of review applications were upheld, but in 2025/26 only 26.8% were. That this is down to AI, however, is merely a guess by the website Resultsense.

Countermeasures: from word limits to ID

Public bodies are already responding, and on both sides of the counter: - Volume and format rules: According to the Guardian, some UK ombudsmen are using word limits and online forms that restrict attachments. The Welsh council Rhondda Cynon Taf is planning a form that asks what outcome the complainant wants. There, stage-two complaints in social services rose from 4 to 20. One MP's office no longer responds to automated “click-to-email” campaigns. - Phone instead of paper: The FOS is calling consumers again to pin down the actual point of dispute. - AI on the agency side: According to the Guardian, Wiltshire Council is looking at AI for summaries and initial responses. According to PPC Land, Hamburg's data protection commissioner has the chatbot LLMoin, built on OpenAI, draft reply letters, but it isn't allowed to make decisions. Complaints there rose 62% to 4,220 in 2025. In the German justice system, assistants such as OLGA (Stuttgart Higher Regional Court), MAKI (Lower Saxony) and FRAUKE and FRIDA (Hesse) are in use. Brazil's labour courts use a system called Galileu. A court there recently sanctioned lawyers who had hidden an instruction to the court's AI inside a filing. That shows how quickly such systems become a target. - Identity and law: Schmitz and his colleagues recommend linking particularly vulnerable services to digital identity credentials. That would allow volume limits per person and pre-filling of known data. They also want agencies to get legal clarity on which countermeasures are permissible at all. According to the paper, the quickest fixes, such as fees, undermine equal access. - Staffing: The Berlin Labour Court is hiring new judges. The insurance ombudsman is asking insurers for more money and asking applicants not to inquire about the status of their case.

Experts warn against shortcuts. A UK practitioner writing in Inside Housing says that anyone who dismisses a complaint as spam because it sounds machine-written risks a finding of maladministration. What counts is the content, not the author. Law firm Sharpe Pritchard stresses that word limits can be proportionate, but there is “no one-size-fits-all solution”. Schmitz fears agencies could shut down digital channels or let courts dismiss more cases without giving reasons. Freedman foresees chatbots ending up talking to chatbots.

The bigger picture

The record numbers are real. How they're being interpreted often isn't. In none of the German cases we examined has a body measured how many submissions are AI-assisted. Almost everywhere, there are plausible other drivers: job cuts, Bürgergeld disputes, flight chaos, new laws. The claim that “AI is causing the flood” is so far an assumption. What is documented is a different shift: longer, legal-sounding and sometimes error-ridden letters that take more review time per case.

Our take on the central question: both things are true at once. The CFPB data suggests AI helps citizens assert their rights. If chatbots steer people to the right arbitration body, that's a win to begin with. But the arms race is already underway: AI writes the complaint, AI summarizes it, AI drafts the reply. Wherever word limits, fees or ID requirements emerge as a result, they hit everyone, including people who have never used a chatbot. Whether more people end up getting justice or access gets narrower depends less on the technology than on whether agencies get the staff and good digital forms they need. An honest debate would above all need one thing that's missing almost everywhere: a systematic count of how many submissions are actually AI-assisted.