I once tried using AI to help write an opinion piece. I wanted to know whether there were any scientific studies that confirmed or disproved certain hypotheses. The AI wrote me beautiful answers in fluent English, complete with citations. The answers were full of all sorts of interesting claims.

I then started asking one by one: what is the source of this specific claim? Sometimes the answer was: "Sorry, no source actually says this exactly. My mistake, thank you for pointing it out." Other times the AI gave a citation, but when I checked it, it turned out the source did not support the conclusion at all.

When I pointed out that the claim did not match the source, the AI politely apologized again and thanked me for drawing attention to its error. After several rounds of this back‑and‑forth, I began to feel that I was helping the AI, not the other way around. I felt like the AI's teacher or coach, pointing out its shortcomings.

Of course, AI tools vary greatly in quality, and you shouldn't expect top‑tier service for a bargain price. I've had the chance to test a more advanced, law‑focused AI solution whose citations and analysis genuinely impressed me. So I'm not trying to claim that AI cannot be useful in law — it certainly can; just like a weapon can be useful for self‑defense, provided you don't shoot yourself in the foot through poor handling.

Regardless of price or quality, you must always be source‑critical when using AI. That doesn't just mean checking whether a source exists, but also diving into its content. AI can help find sources, provide ideas and inspiration, and give feedback on work already done, but it cannot — at least not yet — replace human thinking.

For a long time now, especially from the United States, we've seen news about AI hallucinations ending up in court filings. The problem is much worse there because their legal system is based on precedent.

In a precedent‑based system, judges create new legal rules through their decisions. US law textbooks are often built around case analysis because the rules arise from court practice. In such a system, AI hallucinations have far more room to cause trouble. But even here in Estonia, examples are quietly emerging of AI causing problems in court proceedings.

After Midsummer, the Supreme Court announced (link in Estonian) that at least one such AI‑related case has now reached a final resolution. A group of people had filed a complaint in administrative court to challenge a logging permit. They submitted a document citing scientific studies on clear‑cutting impacts. According to the Supreme Court's press release, the court discovered that the Estonian researchers mentioned in the citations did not actually exist, and the described studies could not be found.

In essence, false factual claims had been submitted to the court. The administrative court fined one complainant €150 and nine others €50. The complainants didn't like this. They argued that using AI is not prohibited and they did not intend to mislead the court. Of course it's not prohibited — but the person submitting a document is responsible for its content, regardless of whether they used their neighbor or an AI to help write it.

The administrative court emphasized that if a complainant cites scientific sources, they must be sure those sources actually exist and contain the information claimed in the filing. That seems obvious and logical. But the complainants didn't think so and appealed to the Circuit Court. The Circuit Court agreed with the lower court. The Supreme Court refused to hear the case, and the fines became final.

Rightly so. If a complainant without legal training makes legal mistakes when turning to court, that is completely understandable — applying the law correctly is the court's job, especially in administrative court. But if someone presents factual claims to the court, citing scientific studies that do not exist, that is entirely their own fault.

Tallinn Administrative Court's president Kaupo Kruusvee explained that administrative proceedings do not require legal knowledge from the person — it is enough for them to describe their real‑life problem and desired solution, and the court must help them formulate a proper claim to protect their rights. He added that for the sake of both the participants' and the court's workload, it would be better if people did not submit AI‑generated, confusing documents, but instead turned to the court directly and in their own words.

One thing is when non‑lawyers trust AI too much. Far more frightening, in my view, is when actual lawyers — even attorneys — fall for AI hallucinations.

Earlier this year, the Tartu Circuit Court issued a decision in a criminal case in which the court scolded the defense attorney for submitting an appeal full of errors that were clearly just AI hallucinations. The attorney had cited supposed Supreme Court decisions that do not exist (the case numbers simply do not exist). In addition, the attorney made several claims about Supreme Court rulings that did not match their content.

One amusing example: "According to the appeal, Supreme Court decision No. 3‑1‑1‑67‑14 deals with the seriousness of threats. In reality, that decision concerns operating without an environmental permit, control measurements, traceability of measurement results, and calculating environmental damage." (Tartu Circuit Court decision of 17.02.2026 in case No. 1‑24‑5343, para. 31)

The Circuit Court summarized its devastating analysis by stating that none of the numerous Supreme Court decisions cited by the defense contained the legal positions claimed in the appeal. The court concluded that the text was AI‑generated and the attorney had not bothered to check it.

The attorney had requested state legal aid compensation corresponding to 4.5 working hours. As a sort of "punishment" for submitting an appeal full of AI hallucinations, the circuit court found it justified to compensate only 1.5 hours — three times less.

 I don't know whether the attorney was also called before the Bar Association for this conduct, but I sincerely hope so. Being a member of the Bar should at least theoretically be a mark of quality. If an attorney allows themselves to make such a crude mistake as submitting AI hallucinations in a court filing, the Bar as a professional body should respond with full seriousness.

--

Follow ERR News on Facebook and X and never miss an update!