Polish Court Criticizes Lawyer for Uncritical Use of AI

Poland's Supreme Administrative Court (Naczelny Sąd Administracyjny, NSA) has issued a ruling that directly criticizes a professional legal representative for uncritically relying on artificial intelligence to draft a court filing. The court found that the rulings cited in the complaint either did not exist in the form presented or concerned entirely different cases than the filing's author claimed.

The case began as a tax dispute. The tax authority issued a VAT decision, and the taxpayer requested that its enforcement be suspended for the duration of the appeal proceedings. The Voivodeship Administrative Court in Łódź refused, so the legal representative filed a complaint with the Supreme Administrative Court, arguing that prior NSA case law supported suspension.

The problem was that the rulings cited in the complaint bore little resemblance to reality. When the court checked the cited case numbers, it found that some had been issued on entirely different dates than stated by the representative, others concerned matters unrelated to the suspension-of-enforcement issue, and still others had actually ruled against the very position the lawyer was trying to support. It is a classic pattern of generative AI hallucination, already familiar from similar cases in courts around the world.

What the Court Said

The NSA left no doubt about how it viewed this conduct. In its reasoning, the court stressed that a paid, professional legal representative is obligated to verify every legal basis and every ruling cited before filing a document on a client's behalf. The court also noted that the value of legal services lies precisely in that verification, not in mechanically copying output generated by a tool available to anyone, including people who are not lawyers.

Uncritical use of artificial intelligence to draft court filings by a professional legal representative amounts to a lack of professionalism - Supreme Administrative Court, ruling I FZ 104/26

Why This Is a Precedent

This is one of the first Polish rulings in which an administrative court addresses, directly and in detail, how lawyers use generative AI tools. Earlier cases abroad, mainly in the United States and the United Kingdom, ended with financial penalties for lawyers who cited nonexistent rulings. The Polish court has moved in a similar direction, but within the context of domestic tax law and administrative procedure.

The significance of this ruling extends beyond the VAT case itself. It establishes that responsibility for the content of a court filing rests entirely with the legal representative, regardless of which tool was used to prepare it. False or misattributed citations of case law can, on their own, become grounds for a court to criticize a representative's professionalism, and in extreme cases, grounds for disciplinary liability.

Professional Responsibility in the AI Era

For law firms in Poland, this is a clear signal that using tools such as ChatGPT, Claude, or other language models to draft court filings requires thoroughly verifying every cited provision and ruling. Courts increasingly have the knowledge and tools to quickly check the authenticity of cited case law, which turns unverified use of AI into a professional risk rather than a mere technical oversight.

Broader Regulatory Context

The ruling coincides with the Sejm (the lower house of Poland's parliament) wrapping up work on Poland's AI law, which is set to establish a new oversight body and introduce regulatory sandboxes for companies deploying AI systems. Professional bodies representing lawyers and tax advisors had already flagged the need for guidelines on using generative AI in professional practice, and the NSA ruling could become a reference point in that discussion.

For law firm clients, this also offers a practical takeaway: it is worth asking how and to what extent AI is used in preparing filings for their cases, and what verification mechanisms a given firm applies. The case shows that an error made by a tool can directly translate into losing a case, as happened to the taxpayer described here.

Sources: NSA ruling on uncritical use of AI by a legal representative (rp.pl), Important NSA position on verifying AI-generated content (gazetaprawna.pl).