The European Court of Human Rights reportedly barred a Romanian lawyer for two years after a court filing contained AI-generated precedents that did not exist or were misrepresented. The lesson for lawyers: the representative remains fully responsible and must demonstrably check every cited authority against a primary source before submission.
This is no longer an abstract risk. The ban affects not the AI model but the professional who submitted the filing. Our analysis: anyone deploying AI in legal work may thereby incur a direct litigation and professional risk, not merely a quality problem.
What exactly happened at the ECHR with the barred lawyer?
The Court’s official announcement states that, on 13 November 2026, its President excluded a Romanian lawyer from representing or assisting litigants before the Court for two years after the relevant Chamber found numerous errors in observations submitted by her, including non-existent cases, fictitious paragraphs in genuine judgments and decisions, and irrelevant or contradictory case-law citations. Reporting by heise online on the incident attributed the erroneous citations to AI-generated material.
The pattern is recognisable: a language model produces fluent, convincing-looking references to cases, dates and citations that are not correct. Without independent checking, these find their way unnoticed into an official filing.
On what grounds may the ECHR bar a lawyer from representation?
The power is set out in the Rules of Court of the European Court of Human Rights. Rule 44D provides that the President may, in exceptional circumstances, direct that a lawyer may no longer represent or assist a party, for a fixed or indefinite period. This requires a reasoned decision following a reasoned proposal by the relevant Chamber. Rule 44D also provides for consultation with the relevant Chamber, any government concerned and any bar association concerned in connection with restoring representation rights after exclusion.
It is therefore an exceptional, reasoned power and not an automatic penalty for any use of AI. Our reading is that the relevant distinction lies in the professional diligence of the work submitted, not in the mere use of an AI tool.
What does this ban mean for lawyers and executives who deploy AI?
Our analysis: because the ban affects the representative and not the model, a lawyer must in practice assume that the reliability of the cited authorities is attributed to their professional responsibility; therefore every firm must establish that generating with a language model is never equivalent to legal research and that verification is a separate, mandatory step before submission. Because in this specific ECHR case a pattern of fabricated or seriously misrepresented references contributed to a two-year ban, this affects not only the handling lawyer but also the partner and firm leadership who set the policy; therefore, in our analysis, the allocation of responsibility should be set down in writing before AI is permitted in case files. Because some errors may only come to light after submission to a judicial body, in our analysis it is prudent to have a competent second reviewer check high-risk filings before they go out the door. Because the cited guidelines from courts and court organisations emphasise independent verification and documentation, it is prudent for the professional to be able to show what they have checked; therefore, in our analysis, a firm should keep a limited verification trail recording the source version, the reviewer, the date and any discrepancies found.
This directly concerns AI in professional practice and the responsibility of professionals and ties in with earlier case law on BGH on AI revision briefs without demonstrable lawyer review. Likewise, law firms adapting their working methods for AI would do well not to cut the checking step.
How do I check and document AI-generated case law before submission?
Guidelines from judicial bodies give a concrete answer here. An official practice guide from the National Center for State Courts advises lawyers to test AI-generated legal authorities directly against primary sources and to document checks and corrections. The practice guide from the National Center for State Courts warns about fabricated cases, statutory provisions and authorities and advises testing every citation, case and claim directly against primary sources, with checklists, multiple review rounds and immediate correction when a hallucination is discovered after submission.
A workable check per cited authority, in the spirit of these official guides and comparable to verifying AI answers by splitting them into sub-claims:
- Does the case really exist, and are the court, the date and the case number correct?
- Can the quoted passage be found verbatim in the official text?
- What was the actual holding, and is the ruling still valid and not overturned?
- Does the source support the specific legal point for which it is cited?
- Was the check carried out via a primary source, not via an AI summary?
- For high-risk filings, has a second competent reviewer looked at it?
In addition, record a brief verification trail: which source version was consulted, who checked, on which date, and which discrepancies were found and corrected. If an error is only discovered after submission, correct it and report it immediately to the body. This trail is not mere bureaucracy: it documents which checks the professional carried out and can therefore be of significance when, after an incident, the care with which the filing was prepared is assessed.
Sources and references
Sources: The article relies on reporting by heise online about the ECHR incident and on official documents from the European Court of Human Rights and the National Center for State Courts.