You must take responsibility for what the algorithm produces
In a Singapore courtroom, a man seeking protection from his wife found himself instead rebuked by the court he had turned to for refuge. His claims of prolonged abuse and destitution collapsed under the weight of contradictory evidence, but it was the citation of a legal case that never existed — conjured by an AI tool he had used to draft his arguments — that drew the magistrate's sharpest attention. The episode places a quiet but urgent question before modern courts: when a machine writes the words and a human signs the page, who bears the burden of truth?
- A man's bid to exclude his wife from their home unraveled when the magistrate found his accounts of malnourishment and physical ruin impossible to reconcile with his own testimony about hauling fifteen kilograms of flour more than a kilometer on foot.
- Buried in his court filings was a citation to a legal case that does not exist — a hallucination produced by generative AI that the husband either failed to verify or chose not to question before submitting it as legal authority.
- The documents themselves betrayed their origins: shifting between first and third person within the same statement, the syntax and structure carried the unmistakable texture of algorithmic composition reviewed by no careful human hand.
- The magistrate dismissed the application, ordered the husband to pay his wife ten thousand dollars in costs, and declined to place any reliance on statements he considered unreliable in both origin and content.
- The ruling landed as a judicial warning to all self-represented litigants: AI may be used, but the responsibility for what it produces — including its inventions — belongs entirely to the person who submits it to the court.
A Singapore man came to court with a detailed account of suffering — years of emotional control, food deprivation, and physical deterioration at his wife's hands, culminating in rotting teeth and brittle bones. He sought a protection order and asked that she be excluded from their shared home. The magistrate listened carefully, then dismissed the case entirely and ordered the man to pay his wife ten thousand dollars.
The credibility of his claims had already frayed before the question of AI arose. The husband had testified to regularly carrying fifteen kilograms of flour more than a kilometer on foot — a detail the magistrate found impossible to reconcile with a body he claimed had been broken by years of malnourishment. Video evidence showed him composed and defiant, not diminished. The magistrate concluded that the man's hardships traced back to a retrenchment in 2009 and his own difficulty adapting, not to his wife's conduct.
But the more striking problem emerged from the documents themselves. In arguing that his wife's counseling records should be disclosed, the husband cited a legal case that did not exist. When pressed, he acknowledged using generative AI to draft initial documents and pull affidavit templates, but denied using it to fabricate evidence. The magistrate was unconvinced. The filings had been prepared with suspicious speed, and the writing shifted between first and third person within the same statements — hallmarks of algorithmic text that had not been carefully read before submission.
The magistrate's response was measured but firm. Litigants are not barred from using AI, he wrote, but they cannot outsource their responsibility to the machine. Whatever the algorithm produces, the person who signs the document must verify it, must think through it, must own it. The husband had done none of that. In submitting a fabricated citation to the court, he had not only destroyed his own credibility — he had tested the integrity of the process itself.
A Singapore man walked into court with a stack of documents he said proved his wife had reduced him to destitution—starving him, controlling him, breaking his body down over more than a decade. The magistrate listened. Then he dismissed the case, ordered the man to pay his wife ten thousand dollars, and issued a rebuke that cut to something larger than one failed lawsuit: the man had used artificial intelligence to write his legal arguments, and in doing so, he had cited a case that did not exist.
The husband had filed for a personal protection order and a domestic exclusion order, seeking to bar his wife from their home. He claimed she had subjected him to emotional and psychological abuse, that she had denied him food and money, that her control had left him malnourished and physically broken. His teeth had rotted so badly he had to travel to India for dental work. His bones had become brittle. It was only after he applied to the Ministry of Social and Family Development in 2024, he said, that he managed to escape what he called her yoke and begin to rebuild his life.
Magistrate Soh Kian Peng read the testimony and found it difficult to square with the evidence in front of him. The husband had testified that he would regularly buy three bags of flour, each weighing five kilograms, and carry them 1.2 kilometers home on foot. The magistrate noted the contradiction plainly: a man so physically devastated that his bones had become brittle, so malnourished that his teeth had rotted away, would not have the strength to haul that much weight that far. The human body has resilience, the magistrate acknowledged, but there are limits to what it can endure. Two videos submitted to the court showed the couple interacting, and in them, the husband appeared composed and defiant—not the picture of a broken man.
The magistrate also considered the husband's history. A retrenchment in 2009 had hit him hard. Rather than finding healthy ways to cope, the magistrate wrote, the husband had fallen into a destructive spiral. The destitute state the man found himself in was not the result of his wife's actions, the magistrate concluded, but of his own inability to adjust and move forward. The application was dismissed.
But there was another problem, one that the magistrate felt compelled to address directly. In arguing that his wife's counseling reports should be disclosed to him, the husband had cited a legal case—one that turned out to be entirely fabricated. When the magistrate asked which artificial intelligence tools he had used and how, the husband denied using AI to create, alter, or fabricate evidence. He said he had used it only to draft initial versions of legal documents and to pull standard templates for affidavits, which he then reviewed and revised.
The magistrate was skeptical. The speed at which the court statements had been prepared was suspicious. More tellingly, the documents themselves bore the fingerprints of algorithmic composition. The husband shifted between referring to himself in the first person and the third person within the same statements. The structure and syntax were peculiar in ways that suggested they had not been carefully reviewed by a human mind before submission. The magistrate placed no reliance on these statements.
What followed was a warning dressed as judicial instruction. The magistrate made clear that litigants are not forbidden from using generative AI. But they bear responsibility for what the algorithm produces, especially when that output becomes evidence presented to a court. The temptation, he noted, is to trust the machine, to let the algorithm do the thinking. That temptation must be resisted. A person who submits AI-generated documents to a court must apply their own mind to what has been written, must verify it, must own it. The husband had failed to do that. He had submitted statements he had not carefully reviewed, cited a case that did not exist, and in doing so, had undermined not just his own credibility but the integrity of the process itself.
Notable Quotes
If he was indeed destitute and so physically weak that his bones had become brittle and his teeth had rotted away, it was quite unlikely he would have been able to haul that quantity of flour over that distance all the way home.— Magistrate Soh Kian Peng
Litigants are not prohibited from using generative AI. But they must take responsibility for its output, especially where this is evidence presented to the court.— Magistrate Soh Kian Peng
The Hearth Conversation Another angle on the story
Why did the magistrate focus so much on the flour? That seems like a small detail.
It wasn't small to him. It was the hinge on which the entire narrative turned. The man was claiming he was so physically destroyed that his body had failed him. But the evidence showed he could still do hard physical labor. That contradiction mattered more than any single claim.
And the AI documents—was the magistrate saying the man deliberately lied, or that he was careless?
The magistrate seemed to be saying both. The man denied using AI to fabricate evidence, but the documents themselves looked machine-made. The real issue wasn't whether he intended to deceive—it was that he submitted something to court without actually reading it carefully. He let the algorithm do the work and didn't check the output.
So the fake case citation—was that the AI's fault or his?
That's the question the magistrate was wrestling with. The man blamed the AI, implicitly. But the magistrate said: you submitted it. You're responsible. You should have caught it. That's the harder standard, and it's the one that has to hold.
What happens to him now?
He lost the case, pays ten thousand dollars to his wife, and his credibility in court is damaged. But the larger question is whether courts will develop clearer rules about AI use before more cases like this appear.
Do you think he actually believed his own claims about the abuse?
The magistrate seemed to think the man had convinced himself of a narrative that didn't match reality. Retrenchment in 2009, years of struggle, a difficult marriage—those are real. But he'd constructed a story where he was entirely a victim, and the evidence didn't support that.