Polish Supreme Administrative Court Slams Lawyer Over AI-Hallucinated Filing

Poland's Naczelny Sąd Administracyjny (NSA, the Supreme Administrative Court) issued a ruling openly criticizing a professional legal representative's uncritical use of artificial intelligence tools in drafting court filings. The case concerned a complaint against the refusal to suspend enforcement of a VAT tax decision, and the NSA ultimately dismissed the appeal, meaning the taxpayer must pay the tax as assessed by the authority.

What the Filing Revealed

The court found clear traces of AI hallucination in the complaint. The court rulings cited by the representative either did not match their stated issue dates at all, or concerned entirely different cases unrelated to the procedure for suspending enforcement of a decision. As Judge Sylwester Marciniak noted, none of the cited rulings reflected the actual position of the cassation court, and some of the quoted judgments were in fact unfavorable to the very party invoking them.

In other words, the representative not only failed to verify the content generated by the language model, but the resulting filing contained arguments that effectively worked against his own client. The court had no doubt the document was produced with AI assistance, since the errors bore the hallmarks typical of generative language models: plausible sounding but nonexistent or distorted legal citations.

The Court's Stance on Professional Ethics

The NSA stressed that it views very critically a situation in which a professional legal representative charges a fee for legal services while the client has every right to expect that filings submitted on their behalf meet the highest standards of professionalism. The court noted that it is hard to speak of professionalism when a paid representative effectively uses tools equally available to people without legal qualifications, without adding any substantive analysis of their own.

This is one of the first such unambiguous rulings by a higher Polish court on a phenomenon that has been growing in the legal community for months. The Krajowa Izba Radców Prawnych (National Chamber of Legal Advisers) already issued recommendations on responsible use of AI in legal practice back in May last year, but the NSA ruling shows that violating these standards can carry real procedural consequences for the client, not just reputational ones for the representative.

Implications for Legal Practice in Poland

The case comes at a time when more and more Polish law firms are using language models to draft preliminary filings, search case law, or analyze documents. The NSA ruling clearly signals that responsibility for the content of a court filing always rests with the representative, regardless of what tool was used to prepare it. Verifying cited rulings and provisions remains a lawyer's obligation, not an optional step.

For law firm clients, this translates into concrete financial and procedural risk, as this case shows: an AI-generated error that went unverified by the representative led to the complaint being dismissed and the unfavorable tax decision being upheld. Legal experts cited in connection with the case point out that similar incidents have already occurred abroad, including in Italy, where a labor court in Turin imposed financial sanctions on a representative for a chaotic filing based on unverified AI citations.

The NSA ruling does not ban the use of AI tools in legal practice, but draws a clear line between assisting a lawyer's work and replacing their analysis. In practice, this means law firms should treat content generated by language models as a starting point for further verification, not a finished product ready to be filed with a court.

Sources: Ważne stanowisko NSA. Profesjonalny pełnomocnik powinien weryfikować czy AI nie zmyśla dla niego faktów (gazetaprawna.pl), NSA: Pełnomocnik nie może bezmyślnie korzystać z AI przy sporządzaniu pism (rp.pl).