AI lowers the barrier for lawsuits, courts haven't caught up yet

Have you ever considered that one day, going to court to file a lawsuit might cost so little that all you need to do is open a chat box?

I've looked at recent data from UK Employment Tribunals several times. In March this year, case volumes suddenly jumped by 39%, and the backlog piled up by another 55%. Judges initially thought it was due to economic downturns or new legislation, but they later discovered that the real drivers were ChatGPT and Grok. Employees stopped hiring lawyers and instead let AI draft complaints—free of charge, and generated in bulk.

My first reaction was: isn't this great? The barrier to legal services has long been criticized as too high; ordinary people couldn't afford lawyers, and now AI has torn down that wall. But looking deeper, things aren't that simple.

The problem lies in the quality of AI-generated complaints. Judges complain that files submitted nowadays are filled with "legal hallucinations" and massive amounts of invalid content.

I've seen code hallucinations when using AI—a function can make up three non-existent APIs. Hallucinations in legal documents are even more fatal; AI will solemnly cite precedents that don't exist or provide completely absurd interpretations of statutes. Previously, a self-represented litigant might submit a complaint of one or two pages; now it's easily hundreds of pages, appearing well-organized and heavily cited, but actually full of AI confidently talking nonsense.

This reminds me of a deeper issue. The design philosophy of AI tools is, to some extent, dismantling the barriers to "expertise." Writing resumes, writing code, writing complaints—essentially, these turn outputs that once required professional training into commodities generated with a single sentence. But while barriers lower, the invisible aspects behind expertise—such as judgment, contextual understanding, and risk awareness—do not automatically come down with them.

What gives the UK side an even bigger headache is that new legislation added 25 grounds for appeal, effectively expanding the scope for AI-generated content. Judges' workloads were already saturated; now they have to sift through piles of AI-generated noise to find genuinely valid claims. Judicial resources are public resources; every hundreds-page document that must be carefully reviewed consumes overall judicial efficiency. An employee whose rights are truly harmed may have to compete for a judge's attention against a pile of invalid complaints batch-generated by AI.

I've used AI tools for work myself for a while, starting with GPT-5.6 about three weeks ago, and later trying Kimi K3. Honestly, for certain structured tasks, it does save effort—writing emails, organizing meeting minutes, even making initial proposals shows clear efficiency gains. But every time I use it, I have a habit: I must review the output myself, because I know that what models generate is sometimes organized language to "look right," not reasoning facts to "be right."

The legal field is more dangerous than the tech field. If code doesn't compile, it throws an error; at least you can run AI-generated code to verify it. But for AI-generated complaints, there is no compiler to help check if the legal basis holds. An ordinary person who doesn't understand law, receiving a 500-page AI-generated complaint, has no idea how much of it is hallucination. They see seemingly professional legal analysis and think it's a royal decree, when in reality it might be a pile of waste paper.

From another angle, this also exposes a structural contradiction in AI implementation. AI's production capacity far exceeds the system's digestion capacity. Generating a document costs almost nothing, but reviewing a document, discerning its legal validity, and contesting it in court—these steps haven't lowered in cost at all. When the cost imbalance between the production end and the consumption end is severe, the entire system gets flooded with meaningless output.

I thought of an imperfect analogy. When AI generates spam email, everyone thinks it's just a matter of filtering. But when AI-generated junk complaints flood the courts, they occupy judicial resources and affect those who truly need relief. This is no longer a technical problem; it's a social governance problem.

My own judgment is that in the short term, courts will be forced to introduce AI-assisted review tools for initial screening, but in the long run, we likely need thresholds or verification mechanisms for AI-generated legal documents. It's like selecting technical solutions—we can't just look at generation efficiency but must also consider downstream absorption capacity.

So, if you plan to use AI for legal matters, my advice is: treat it as a search tool, not a lawyer. Let it help you understand statutes and organize thoughts, but for the final document, either have a professional review it or verify each legal basis yourself. For every sentence AI generates, ask: what is the basis for this statement, and can it be verified?

AI can help you write a complaint that looks decent, but courts don't care if it looks decent; they care if it stands up.