AI’s growing pains in the Maltese courts

Artificial intelligence has made its way into Malta’s courts, and not just through the odd submission whose phrasing gives away that friend Claude has been called in to lend a hand. 

This article is inspired by footnote 17 of a judgment I was reporting on a couple of weeks back. I might have skimmed past the tiny print altogether had two words not jumped out at me: Artificial Intelligence. 

Judge Henri Mizzi was handing down judgment on a case involving more than €500,000 in donations by a charitable retired teacher in the years before she was declared legally incapable of managing her own affairs. Over half of the donations had to be refunded. 

One of the charities, NGO Moviment Ġesù fil-Proxxmu, had gone through the lengthy case, which was challenged on several fronts through the years, without ever saying that it was, in fact, a foundation. Yet, in its final submissions, it suddenly described itself as one. 

“In order to state, correctly, that Ġesù fil-Proxxmu is a foundation, it was necessary for evidence of this to be found in the acts,” Judge Mizzi wrote. “The evidence is entirely lacking and, therefore, the submission should never have been made.” 

He continued: “Outside Malta, many lawyers have been found guilty of contempt because, through the use of artificial intelligence, they made improper submissions, often by referring to jurisprudence that the intelligence had invented. This, and several similar cases, is an identical situation.” 

The judge stopped short of calling it an AI hallucination, or finding the lawyer guilty of contempt. Which means Malta is still waiting for a such reckoning but the writing, as they say, is on the wall. 

The day an accused read a definition off ChatGPT 

Magistrate Donatella Frendo Dimech showed rather less patience for former Labour Party Secretary General Jimmy Magro’s attempt to lean on artificial intelligence from the witness stand in a bid to talk his way out of the corruption case against him. 

Magro’s four-year and eight-month prison term for trading in influence was confirmed last April on appeal, ostensibly the first time in recent memory that a former senior political party official has been jailed for corruption. 

Magro, who had solicited a €10,000 payment to ensure a bid would succeed in a public tendering process, had been asked to explain what he meant by the expression “commercial take” in an email sent in 2014. 

Magro sneered at the Commission Against Corruption’s definition of the term, saying it was interpreted as they saw fit from a clapped-out dictionary: A “dizzjunarju tan-19 xejn xejn.” 

He said he had turned to “modern technology” to establish the meaning of “commercial take in business”. The modern technology turned out to be ChatGPT 3.5. 

He then proceeded to read out, word-for-word, the rather robotic answer the chatbot had provided him with into the record: “In business, commercial take typically refers to the revenue or sale generated from commercial activities. It represents the amount of money earned from selling products or services to customers. It is a key metric in assessing the financial performance and success of a business operation.” 

The problem, as Magistrate Frendo Dimech pointed out, was that ChatGPT did not exist at the date of the email. 

“When the accused is asked by his lawyer what he meant by the phrase ‘commercial take’, instead of answering this question frankly, he incredibly refers to ChatGPT,” the magistrate wrote, saying that when the term was used on 24 July 2014, “the world had not yet dreamed of it”. 

She added that Magro would have been better off relying on his “sacrosanct right to silence”, since his “forced explanation” had begun to “bury the credibility of his testimony”. 

AI as the bench’s research assistant 

Artificial intelligence has not always featured as the villain. In fact, our bench has turned to it as a research tool on several occasions.  

Last April, when self-styled ex-gay Matthew Grech and the two PM news co-hosts were acquitted of the charge of advertising gay conversion therapy, Magistrate Monica Vella’s footnotes cite a Microsoft Copilot URL as a research source for working out how the conversion therapy is regulated across different jurisdictions. 

Other judges prefer Gemini. 

Last February, when deciding on the compensation that should be given to the owners of a Balluta apartment under Malta’s old rent-control laws regime, Judge Giovanni Grixti referred to GeminiAI to obtain 1987 rental figures. 

In a footnote, Gemini, citing the National Statistics Office, the National Minimum Wage National Standard Order, and reports by MaltaToday and The Malta Independent, duly turned up the fact that in in 1987, the minimum wage was €69.60 a week, or €3,617 annually. The estimated rental value of the property stood at €12,379.  

The court, working through the figures, wondered aloud how realistic it really was for a landlord to have charged “four times the annual minimum wage of a Maltese worker” for the property. 

Noting that the average home back then cost a mere €55,000, the judge knocked back a further 10% off the compensation to account for the discrepancy, bringing the final figure to €53,114. 

Before the chatbot era 

The term “artificial intelligence” has not cropped up that extensively before the Maltese courts, unlike its international counterparts, where it’s increasingly becoming a headache. 

The earliest reference for artificial intelligence identified for this article predates AI as we know it today by well over a decade. In 2010, a Court of Superior Appeal dealt with a navigation system that was described as using “artificial intelligence” in an intellectual property tussle. 

By this, the court meant nothing more exotic than its ability to respond to user input.  

It was, in other words, a very different kind of artificial intelligence from the technology now appearing in courtrooms. 

AI generated pornography 

Not every case makes for light reading. In a 2025 judgment, the court dealt with an AI-generated pornographic image purported to depict a child under nine years of age. 

The accused, who received a probation order, explained that the original photograph—taken from Facebook—showed a person who was not a minor, and that the AI generated image, which was of a non-sexual nature, had been produced “just out of curiosity”. 

Having personally examined the material, the court concluded that while it warranted punishment, it fell short of the threshold for imprisonment. 

The case, decided a year ago, leaves the legislator with some homework to do: How ought the criminal law treat imagery that never involved a real photograph of a real child, once artificial intelligence is doing the generating? 

ChatGPT also managed to cause a stir at the Maksar trial last year. A reserve juror was found to have smuggled in a tablet, which he had been using, by his own account, “to see about the situation of the trial”.  

He was fined €1,000 and removed from jury duty; the trial itself carried on regardless. 

Which brings us to the irony of it all. 

Malta is, after all, encouraging its citizens to get chatty with chatbots. Back in May, Malta became the first in the world to hand out free year-long subscriptions to ChatGPT Plus or Microsoft Copilot to any citizen who completes its AI For All course. 

This scheme is championed by Economy Ministry led by Silvio Schembri, in partnership with OpenAI, Microsoft and the University of Malta.  

The technology, in other words, is being ushered into everyday Maltese life at precisely the moment our legislators, law enforcers, judges, magistrates and lawyers are still working out where, if anywhere, it belongs within the legal system. 

In dubio pro reo 

AI, when it isn’t off hallucinating, simply does as it’s told, and has no stake in misleading a court. But it’s worth asking what it actually means for a machine to be asked anything at all—even granting it, for argument’s sake, something resembling thought. Whatever it’s doing up there, it cannot, in any meaningful sense, doubt. 

René Descartes’s famous cogito, ergo sum (I think, therefore I am) tends to get rather remembered as starting with thought. But as Anne Carson recently reminded us, it started with doubt. 

Descartes set about stripping away every belief that could conceivably be questioned—his senses, the physical world, even mathematics itself—until the only thing left standing was the fact that someone was doing the questioning.  

Thinking only got him to existence because doubting got him there first, which is presumably why the fuller, later gloss on his logic runs dubito, ergo cogito, ergo sum (I doubt, therefore I think, therefore I am). 

A hallucinating chatbot never doubts. It doesn’t hesitate over a fabricated case citation and wonder, ‘hang on, is this right?’.  

In a criminal courtroom, a lot is built on beyond reasonable doubt. And it is probably why a courtroom is the most dangerous place for AI to hallucinate—research tool or not. 

While AI does some of the thinking, but none of the doubting, the people doing the asking can remain very much on the hook for what gets submitted, said, and done in their name— doubt included. 

Perhaps the time is ripe for us to start working out what responsibility looks like on that front.