“A thing is not necessarily true because a man dies for it.”
—Oscar Wilde
I am hardly a trained professional in the practice of law, but it doesn’t take a lawyer to know that when you sign your name, it is wise to know the truth of what you’re signing – at least enough to stake your reputation, if not your life, on it. A thing may not necessarily be true because a man dies for it, but men do tend to hew to the truth when their name is on the line.
Even the man who hides an awful truth with silence or lies must still carefully attend to the truth he hopes to conceal. The best liars never assume what is true, nor do the most honest of men. The enemy of liars and honest men alike is a kind of creeping carelessness when it comes to the truth.
And though I cannot prove it, I suspect most people fall into this lukewarm middle way most of the time – neither always honest, nor always lying – but simply trained to float free in a haze of assumptions, probabilities and plausibilities. Most people are on autopilot in relation to the truth, myself included, especially when the stakes are low and the day seems routine and boring.
Yet, even when the stakes are high, the most highly-trained minds can still succumb to the carelessness of habit – and sign their name to a false shortcut they assumed to be true.
That’s what I believe happened to three Alabama attorneys recently sanctioned for their use of ChatGPT to create court filings, which contained hallucinated legal citations that simply don’t exist.
As recently reported by 1819 News’ Caleb Taylor:
Northern District of Alabama U.S. District Judge Anna Manasco sanctioned three Butler Snow attorneys on Wednesday over their role in submitting court filings created by ChatGPT in litigation defending the Alabama Department of Corrections.
The attorneys had previously apologized to the court and described the episode as an ‘isolated event.’
Manasco publicly reprimanded Matthew Reeves, William Cranford, and William Lunsford. Lunsford is a contract deputy Alabama Attorney General, who has been paid tens of millions of dollars to defend the state in various lawsuits against the Department of Corrections. Manasco disqualified the trio from further participation in the case.
A copy of Manasco's order on Wednesday will be sent to the Alabama State Bar ‘for further proceedings as appropriate.’
Manasco said in her order, ‘Fabricating legal authority is serious misconduct that demands a serious sanction. In the court's view, it demands substantially greater accountability than the reprimands and modest fines that have become common as courts confront this form of AI misuse.’
Indeed, “AI misuse” has become increasingly common in the legal profession.
Another Alabama lawyer, James Johnson, was also recently caught using AI in federal court filings while defending an alleged drug kingpin in a high-stakes case.
A similar incident occurred in a New York federal court in 2023.
In 2025 alone, the careless use of AI in court filings occurred in Utah, Wyoming, and a Colorado defamation case involving two attorneys representing MyPillow CEO Mike Lindell. This online database claims at least 231 legal decisions stretching back to 2023 involving AI hallucinations in courts around the world.
So why do these generative AI models hallucinate anyway?
Because, like most humans, they too are lukewarm when it comes to the truth.
Most large language models (LLMs) produce answers based on the most probable or plausible next word sequence after being trained to recognize common patterns and associations. They’re essentially masters of echoing and reflecting human consensus and convention rather than oracles full of veracity or independent truth.
In the context of a legal briefing, that means an LLM can easily hallucinate a citation that is completely made up as long as it sounds plausible in the larger context of the briefing.
Indeed, having recently used LLMs myself (I use the latest version of Grok these days) to analyze large collections of my own writing, I have witnessed these hallucinations firsthand. Grok 4 is great, but on occasion, when faced with a fairly complex and large dataset (tens of thousands of words all at once), it will respond with perfect confidence that I wrote something I never did – going on-and-on attributing quotes to me that I never put to the page. Sometimes the quotes plausibly sound like me, but they aren’t quite me.
In a way, it’s frustrating, yet also a relief to see these hallucinations. It’s a reminder that LLMs are just a tool for humans, not a replacement – though it is also unnerving to realize how much humans are just like LLMs in their day-to-day lives.
So much of human interaction isn’t about saying what is true, but about saying what is expected – a consensus or convention reflected, a trained response.
When someone asks, “How are you?” most people know they shouldn’t respond with the deepest truths of their inner psyche. They should just say “fine,” or “been better,” with a “How are you?” in response.
Most low-stakes human interactions are like that, trained answers to trained questions. The same drive home doesn’t require mapping the route anew every day when the trained autopilot of our perception can handle the way just fine.
That’s the irony of the three Alabama lawyers recently sanctioned for using LLMs in their court filings – that they acted too much like LLMs themselves in looking for a shortcut they assumed to be true. Yet, unlike any Artificial Intelligence, those men still had to sign their names attesting to the truth in a case that would affect the life of a man in prison. They had to bear the responsibility of their words, staking their reputations to what they believed to be true – and despite their professional training, they failed to recognize the stakes.
For now, no AI can take the same responsibility for its own words. For now, no AI can die for the truth. Maybe that’s a good thing, as I suspect an AI that could die for the truth could lie for it, too.
A thing may not necessarily be true because a man dies for it, but sometimes the discovery of truth means a man must stake his life on it.
The truth found in improbable and implausible places, the truth in defiance of consensus and convention, the truth of an inspired and intrepid individual standing against the rest of the world – only those who can take responsibility could ever discover such a thing.
Joey Clark is a native Alabamian and is currently the host of the radio program News and Views on News Talk 93.1 FM WACV out of Montgomery, AL, M-F 12 p.m. - 3 p.m. His column appears every Tuesday in 1819 News. To contact Joey for media or speaking appearances, as well as any feedback, please email [email protected]. Follow him on X @TheJoeyClark or watch the radio show livestream.
The views and opinions expressed here are those of the author and do not necessarily reflect the policy or position of 1819 News. To comment, please send an email with your name and contact information to [email protected].
Don’t miss out! Subscribe to our newsletter and get our top stories every weekday morning.
