48 Non-Existent Rulings: What the Canary Islands High Court's Order Requires from Your Firm

Citing AI-generated jurisprudence without verification can cost a lawyer a penalty and disciplinary action from the bar association: On February 10, 2026, the Tribunal Superior de Justicia de Canarias fined a lawyer 420 euros for including 48 non-existent rulings in his brief, which were suggested by a general-purpose AI chat, and reported the matter to the Colegio de Abogados. The court order did something even more unprecedented: to calculate the penalty, it referenced the price of legal AI tools that would have prevented the error ("starting from 70 euros per month"), as stated in the official note from the Poder Judicial. In this article: what that court order exactly states, why nearly 90% of documented incidents worldwide affect solo practitioners or firms with fewer than 25 lawyers, and the three — free — rules for working with AI this week without jeopardizing your practice.

And there's something even worse than the fine. The sanctioned individual himself argued in his defense the damage he was already suffering: the comments and mockery in lawyers' group chats. The €420 is paid once. Reputation in a small, connected community is paid for years.

It's not an isolated case. And it doesn't happen to the big players.

That same year, the High Court of Justice of Galicia found 24 fabricated citations in a brief — and opened disciplinary proceedings for procedural bad faith. In Navarre, another similar case was dismissed because the lawyer apologized in time. Same error, different outcomes: the difference was marked by how it was managed.

Update (Jul 15, 2026) · The TSXG has resolved the outstanding matter from this series: a €1,800 fine for procedural bad faith (Art. 75.4 LJS) and referral to the Bar Association of A Coruña, citing the Canary Islands precedent and the "humanity reserve" from the CGPJ doctrine (Instruction 2/2026) — AI can help you prepare and study; verification is not delegated. The lesson remains unchanged: how errors are managed continues to make all the difference.

And the global picture confirms the pattern. The international database tracking these incidents (maintained by a researcher from HEC Paris) documented 1,227 cases in April 2026. The important data point: almost 90% is concentrated in lawyers practicing alone or in firms with fewer than 25 attorneys.

This is not a problem for large firms, which have entire departments reviewing. It's a problem for the small firm. Yours, your partner's, the one on the corner.

Your firm is an SME. But it's not just any SME.

I've been writing about SME digitalization for years, and there's something that law firms and consultancies have that no other SME has. Three things, actually:

1 · Your files aren't just data: they're others' secrets. The store that digitizes its inventory risks its information. You, when uploading a file to any tool, risk your client's information — and professional confidentiality makes no distinction between 'I uploaded it to the cloud' and 'I told it at a bar'.

2 · Responsibility cannot be delegated. Ethically, you are accountable for what your firm signs. Always. AI does not face disciplinary action; you do.

3 · Your raw material is text — exactly where AI shines and where it fails most elegantly. An AI hallucination doesn't appear to be an error. It looks like a judgment: with its number, its date, and its courtroom. The most cited academic research on the topic (Stanford University, published in a peer-reviewed journal) found that even specialized paid legal tools, which significantly reduce error compared to general-purpose chatbots, do not eliminate it.

That's why, when someone sells you an AI "without hallucinations, guaranteed," you know what they're doing: blowing smoke. The serious question isn't "can it fail?" — they all can. It's "how much does it cost me to verify it, and is there a record that I did?"

The Three Rules (Free This Week)

You don't need to buy anything to start working better tomorrow:

Rule 1 · No citation without the source upfront. If the tool provides you with a ruling, that ruling exists when you see it in CENDOJ — not before. This is exactly what the Canary Islands ruling criticized for not having done. If verifying a citation takes you twenty minutes, the problem isn't you: it's the tool.

Rule 2 · Client files are not uploaded. Work with AI on templates, outlines, and anonymized texts. Client files do not enter a generalist chat, period. Digitalizing your firm doesn't require compromising professional secrecy; anyone who tells you otherwise hasn't understood your profession.

Rule 3 · Document your due diligence. Note what you verified, where, and when — even if it's just in a table. The day something goes wrong (and in this line of work, something always does), your record, not your memory, will distinguish an excusable error from negligence.

And a clarification I owe you, because there's a lot of noise circulating: no, Europe won't "fine you in August." What is applicable from August 2, 2026, are transparency obligations; the high-risk regime of the AI Regulation is postponed until December 2027, and a co-pilot used in a private firm is not automatically "high risk." Anyone using that date to rush you is selling to you with fear.

This Newsletter Is Changing (Without Changing Its Name)

Pyme Digital PRO was born to help SMEs adapt to the digital environment. That continues. What is narrowing is the focus: from today, this newsletter focuses on the professional SME — the law firm, the administrative consultancy, the business that thrives on its expertise and its signature — because it is the SME with the least margin for error of all, and the worst served by the generic discourse of "digital transformation."

Every two weeks: a real case with its source, a step-by-step applicable method, or regulation explained without alarmism. All verifiable. If you subscribed expecting something else and this doesn't serve you, unsubscribing is fine — I prefer two hundred readers whom I truly serve than two thousand whose time I waste. No fluff also works internally.

What I'm Building (And Why I'm Telling You Here)

I am a jurist by training and have been managing European funds for over thirty years, seeing, from the inside, why digitalization strategies fail to be implemented in practice. Now I am working with a small group of law firms and administrative consultancies on something very specific: an AI-powered work system in which every statement is anchored to its official source and due diligence leaves a record. It's not finished — I'm building it, and that's precisely why I'm telling you about it here: this newsletter will document the process, with its metrics and errors, openly.

Frequently Asked Questions

Can a Spanish Court Sanction for Citing AI-Generated Rulings?

Yes. The High Court of Justice of the Canary Islands fined a lawyer €420 in February 2026 for citing 48 non-existent resolutions suggested by a general-purpose AI, with a referral to the Bar Association. The High Court of Justice of Galicia opened disciplinary proceedings for 24 fabricated citations; in Navarre, a similar case was dismissed after the lawyer's apology. The difference in outcomes was determined by how the error was managed.

What is an AI 'hallucination' in the legal field?

It's false information generated by AI that appears true: a fabricated judgment with a plausible resolution number, date, court, and reporting judge. It doesn't look like an error; it looks like a real citation. That's why the only reliable defense is to verify each citation against the official source (in Spain, CENDOJ) before signing off.

Do paid legal AI tools eliminate hallucinations?

No. The reference academic study (Stanford University, published in the Journal of Empirical Legal Studies, 2025) found that specialized legal tools substantially reduce errors compared to generalist chatbots, but do not eliminate them. Any provider guaranteeing 'zero hallucinations' is promising something not supported by published evidence.

Does the AI Act impose obligations on me in August 2026 if I have a small law firm?

What is applicable from August 2, 2026, are transparency obligations; high-risk obligations (Annex III) are reserved for December 2027, and an internal private law firm copilot is not automatically 'high-risk.' Anyone telling you that 'everyone will be fined in August' is selling fear, not compliance.

How do I verify if a judgment cited by AI actually exists?

By searching for it in CENDOJ, the official jurisprudence search engine of the General Council of the Judiciary. If the resolution doesn't appear there, it doesn't exist for the purposes of your brief. This is exactly the verification that the order from the High Court of Justice of the Canary Islands criticized for not having been done — and it provides proof of your due diligence.

Why does this affect small law firms more than large ones?

The international incident database (D. Charlotin, HEC Paris) documented 1,227 cases in April 2026, with nearly 90% concentrated among solo practitioners or firms with fewer than 25 lawyers. Large firms have entire departments for review; in small firms, verification depends on a single person and their method.

Sources for this article

• Official statement from the Poder Judicial regarding the TSJ de Canarias ruling (Feb-2026).
• Magesh et al., Journal of Empirical Legal Studies (Stanford University, 2025)
• D. Charlotin's incident database (HEC Paris, Apr-2026).
• Regulation (EU) 2024/1689 and Omnibus package (Jun-2026).

Sources verified on Jul-6-2026. This blog doesn't ask for faith: it links to the source.

Every two weeks I publish an analysis like this in the PYME Digital PRO newsletter: digitalization for law firms, consultancies, and professional SMEs, without the hype. Subscribe on LinkedIn.