Here’s a story you may have heard, and if you haven’t, it’s a good one.
In the 1960s, a teenage boy named Frank Abagnale ran one of the greatest con sprees in American history. Before he was old enough to legally buy a drink, he’d talked his way into a Pan Am pilot’s uniform and flew all over the world for free, deadheading in cockpit jump seats on the strength of a fake ID and a confident smile. When that got too hot, he became a doctor, supervising real physicians in a Georgia hospital. Then a lawyer. Then a college professor. Along the way he cashed something like two and a half million dollars in forged cheques across dozens of countries, always one charming step ahead of the law, until they finally caught up with him in France when he was twenty-one.
You may have seen the movie, Catch Me If You Can.
It’s a hell of a story. It became a bestselling memoir, then a hit Broadway musical, then a Steven Spielberg movie with Leonardo DiCaprio playing Frank and Tom Hanks playing the FBI man chasing him. And for decades, Frank Abagnale made a very good living telling it over and over again and commanding big fees on the corporate lecture circuit as the reformed genius con artist who now teaches the good guys how the bad guys think.
There’s just one problem. And it’s the best part…
The lie inside the lie
A few years ago an author named Alan Logan got curious and did the one thing forty years of audiences never bothered to do. He checked. In his 2020 book The Greatest Hoax on Earth, Logan lays out years of digging through court records, old newspaper archives, and arrest reports, and he brings in the people who were actually there, including a Delta flight attendant named Paula Parks whom Abagnale had latched onto in 1969, and Mark Zinder, Abagnale’s own former booking agent.
Piece by piece, the documented record they’d uncovered didn’t match the legend.
Some of it falls apart with a single phone call. Abagnale used to tell Johnny Carson and packed lecture halls that he’d once bluffed his way into a job as a sociology professor at Brigham Young University. A local journalist checked that one out the hard way, by simply ringing up the school and asking. The university’s answer, more or less, was who? They’d never heard of him.
So here’s the thing that makes this story truly fascinating. Frank Abagnale’s masterpiece wasn’t the pilot’s uniform or the rest of the yarns. It was the second con, the one stacked right on top of the first. His real genius was convincing the entire planet that his con-man legend was true, and then getting paid to repeat it for forty years while nobody ever stopped to check. He didn’t just tell a lie. He told a lie about being a liar, and then we gave him standing ovations and a movie deal for it.
That’s a deception with a lie folded inside a lie inside a lie. And it worked for one simple reason, which happens to be the same reason your LLM AI fools you each and every single day.
A good lie sounds exactly like the truth
The reason a con like Abagnale’s runs for decades is that a good lie doesn’t sound like a lie. It sounds like the truth in the truth’s own clothes. Right details, right rhythm, right easy confidence. It fits so snugly into what you already expected to hear that the thought of checking it never even occurs to you.
That’s precisely how a large language model works. Underneath the fluent prose, it isn’t tracking what’s true. It’s tracking what sounds right, which word most plausibly comes next based on the mountain of patterns it swallowed in training. And it hands you the real answers and the invented ones in the same smooth voice, with the same steady certainty. There’s no tell. There’s no wink. No blink. No sitting back from the table to look weak. In fact, it’s never more sure when it’s right and less sure when it’s making things up. It’s just always sure, exactly the same way Frank Abagnale was always sure.
This isn’t some rare hiccup, either. It’s the whole design, and the numbers are genuinely hard to believe for something that most of us use and trust every day. Stanford researchers found chatbots hallucinating on somewhere between 58 and 82 percent of legal research questions. Other studies push the fabrication rate on specific legal queries as high as 88 percent. More than eight in ten legal professionals say an AI has handed them fake case law, complete with plausible judges and citations to cases that never existed. One New York lawyer filed a brief built on AI-invented cases and then assured a federal judge they were right there in the databases. They weren’t. Not a single one.
Every one of those is a tiny Abagnale in a borrowed uniform. A story so fluent and well-fitted that nobody thought to check, right until it fell apart in front of someone who mattered.
What it takes to actually check
You can’t fix this by making the machine sound more sure of itself. A slicker sounding con man is just a con man you believe for a longer time. But the whole industry keeps shipping bigger, smoother, more articulate versions of the same thing and then acting shocked when it fibs to them in beautiful, well-punctuated paragraphs.
The fix is a system that actually keeps track of what it knows versus what it’s only reaching for and then tells you the difference instead of blending both into one seamless answer.
That’s what Vertus is built to do, through something it calls Metacognitive Trajectory Analysis. Instead of just generating a fluent answer and letting you assume the fluency means it’s real, the system watches its own certainty the whole way through its reasoning. It notices when it’s standing on solid ground and where it’s just guessing, and it treats that boundary as the single most important thing to hand you, not the thing to smooth over. When it knows, it tells you. When it’s reaching, it tells you that too. And when it can’t back something up, it says so plainly, instead of doing the Abagnale, which is to just say it with a firm handshake and a good suit and let you find out the hard way.
It’s the difference between a source that cares whether the thing’s true and one that only cares whether the thing lands. And when your name’s on the brief, that’s not a small difference. It’s the only one that counts.
Drop it into reality
The unsettling thing about Frank Abagnale was never that he lied. Plenty of people lie. It’s that he lied so smoothly, so plausibly, with such easy confidence, that the truth never got a word in, and a whole civilization of smart people nodded along for the better part of half a century.
We’ve spent three years doing the same thing with our LLM AIs. We’ve been dazzled by how sure they sound and we’ve barely paused to ask whether sounding sure and right have anything to do with each other. They don’t. A machine that sounds brilliant while being wrong most of the time isn’t an assistant. It’s a very well-dressed con man with a great vocabulary and an even better straight face.
So the next time an LLM hands you an answer that sounds just right, do the thing nobody did for Abagnale for forty years. Drop it into reality, and actually check. Ask whether the fellow in the pilot’s uniform is really a pilot.
Vertus was built to know the difference between sounding true and being true. Which, it turns out, has been the whole con all along.
Disclaimer: GeekWire newsroom and editorial staff were not involved in the creation of this content..
