AI floods German courts with 4,500-page filings

AI floods German courts with 4,500-page filings

What happens when a machine can write in seconds what would take a lawyer weeks? The natural filter disappears. In Germany, social and civil courts are seeing an unprecedented surge in lawsuits drafted by artificial intelligence — endless documents packed with legal jargon and technical terms that often cite court rulings that do not exist. A judge recently received an AI-generated legal brief of 4,500 pages. And unlike Joyce's Ulysses, he could not leave it on the shelf unread.

The judge cannot ignore the 4,500 pages

The problem is not that AI is hyper-intelligent. It is that it produces nonsense faster than humans can refute it. In German social courts, where any citizen can sue without a lawyer, almost every claimant now has a free legal assistant: ChatGPT. The AI generates endless reports in seconds, embellished with invented case law. Every document must be reviewed in full, even if it is clear by page 2 that it is nonsense. Because, in theory, there could still be a coherent sentence on the penultimate page.

The result is a measurable collapse. In Baden-Württemberg, urgent procedures in social courts rose by 60% in a single year, and disputes over the citizen's allowance (Bürgergeld, Germany's basic unemployment benefit) increased by up to 80%. In Bavaria, claims related to the same benefit grew by 60% in the first half of 2026. While judges wade through ramblings, cases pile up. People suing pension or long-term care insurers wait month after month. The rule of law promises to resolve matters within a reasonable time. The reality is different.

What the Ministry of Justice proposes

The Federal Minister of Justice is planning a reform of procedural law with rules on the format and length of submitted filings. The idea is not new: in public tenders, clauses limiting the length of bids have been applied for years to avoid documents of hundreds of thousands of pages that no one can read. The CJEU, for example, already has page limits in its rules of procedure. What AI changes is the scale: a 4,500-page filing used to cost weeks of a lawyer's work; now it costs seconds.

Some go further and propose equipping the courts with artificial intelligence. One machine drafts the complaint, another summarizes it, and a third delivers the verdict. The AI judge would then decide. In the name of the people, or in the name of the robots? The temptation to automate justice has its supporters and detractors. The former argue that a judge who hands down rulings that defy common sense but are "legally sound" does not contribute much either. The latter point out that AI does not need to be hyper-intelligent to weaken institutions: it is enough for it to produce nonsense faster than we can refute it.

Invented rulings already have precedents

The problem of fake citations is not theoretical. The best-known case is Mata v. Avianca, in New York in 2023: a lawyer filed a brief citing precedents that ChatGPT had made up, and the judge sanctioned him. Since then, similar cases have occurred in other countries. The rule that is taking hold is simple: whoever signs is responsible for every citation, whether written by them or by a machine. In Germany, the combination of claimants without lawyers and free AI legal assistants multiplies the risk.

The most likely solution involves templates and length limits in proceedings without a lawyer. Problem solved. Or so they say. Because if AI has proven anything, it is that as soon as one door closes, it finds another window through which to slip in 4,000 pages. And someone will have to read them. In the name of the people, of course.