Artificial Intelligence and the Duty of Diligence: The Procedural Risk of Relying on Fictitious Content

Artificial Intelligence and the Duty of Diligence: The Procedural Risk of Relying on Fictitious Content

Generative artificial intelligence has become a useful tool for lawyers, consultants, and legal departments: it helps organize ideas, summarize documents, prepare initial drafts, and identify possible lines of argument. However, its use in professional contexts — and especially in litigation — poses an obvious risk: that the professional may incorporate content that appears plausible but is false. So-called AI “hallucination” is not a minor error when it affects legal sources, case law, doctrine, or relevant facts. It can undermine the professional’s credibility, harm the client, and lead to procedural or disciplinary sanctions.

Real Cases of Sanctions for Citing AI-Invented Sources or Case Law

The leading case is Mata v. Avianca, decided in 2023 by the United States District Court for the Southern District of New York. The claimant’s lawyers filed a brief that included non-existent judicial decisions generated by ChatGPT. The court imposed a $5,000 fine and emphasized that the issue was not the use of AI itself, but the failure to verify the authorities cited before submitting them to the court.

Since then, courts have shown decreasing tolerance for this type of mistake. In February 2025, a federal court sanctioned three Morgan & Morgan lawyers for including fake AI-generated cases in court filings. The judicial response confirms a clear trend: the use of AI does not displace the professional duty of review; rather, it reinforces it.

Similar incidents have also occurred in Canada. In Zhang v. Chen, 2024 BCSC 285, a lawyer was ordered to pay certain costs incurred by the opposing party after including two non-existent cases in family law proceedings. The court did not find proven fraudulent intent, but it did find a failure to meet the minimum standard of diligence required.

Spain is not immune to this phenomenon either. In March 2026, the High Court of Justice of Navarra warned that fraudulent or careless use of AI may be sanctionable as procedural bad faith, although it closed the proceedings against a lawyer who had cited invented case law, taking into account her prompt apology and the absence of bad faith. In the Canary Islands, it was reported that a lawyer was fined €420 after including 48 fake judgments generated by AI in an appeal, with the matter referred to the Bar Association for potential disciplinary review.

What to Verify Before Using AI-Generated Content in a Professional Context

The first rule is simple: no legal citation should be incorporated without independent verification. If AI suggests a judgment, it must be checked in an official or reliable database — such as CENDOJ, the Official State Gazette, EUR-Lex, Curia, professional legal databases, or official judicial repositories — to confirm that the decision exists, that the case number is correct, that the court matches, and that the cited passage accurately reflects the applicable legal doctrine.

The second check concerns context. A real judgment may be misinterpreted, overturned, refer to a different jurisdiction, or not be applicable to the specific case. AI may mix jurisdictions, translate legal concepts inaccurately, or present an isolated ruling as settled case law.

The third verification must focus on facts and data. Dates, names of laws, amounts, procedural deadlines, competent authorities, and documentary references must be reviewed manually. In regulated sectors — data protection, consumer law, competition, intellectual property, advertising, finance, or healthcare — the risk increases because an incorrect statement may have regulatory or contractual consequences.

In addition, confidential information, personal data, or trade secrets should not be entered into an AI tool without an adequate legal, contractual, and technical basis. The duty of diligence requires not only verifying the output, but also controlling what information is introduced into the system.

How to Document the Use of AI to Reduce Legal Exposure

The best defense against a potential allegation is not to state that “AI generated it,” but to demonstrate that there was effective human supervision. To do so, it is advisable to implement an internal AI-use protocol with at least three layers.

First, a usage record: which tool was used, for what purpose, on what date, and for which part of the work. It is not necessary to retain every prompt in every scenario, but relevant uses should be documented when they affect court filings, legal reports, opinions, or professional deliverables.

Second, verification traceability: saving links, screenshots, database identifiers, or official references proving that the cited sources were checked. In court filings, each citation should be reviewed before final submission, and an internal record of that review should be kept.

Third, an approval policy: AI-generated content should not be sent directly to a client, a court, or an administrative authority without review by a responsible professional. AI may assist at the drafting stage, but the legal judgment, selection of sources, and final validation must remain human.

Ultimately, AI does not eliminate the duty of diligence: it transforms it. Used responsibly, it can improve efficiency; used without control, it can turn an apparently solid legal filing into a source of professional liability. The key is not to prohibit AI, but to treat it for what it is: an auxiliary tool, not a source of authority.