ChatGPT Confessed Under the Reid Technique

A renowned criminologist at Penn Law spent a weekend getting ChatGPT to confess to a crime it couldn’t possibly have committed. The bot kept insisting, correctly, that it was architecturally incapable of hacking into his email and sending text messages to his contacts. Professor Paul Heaton was armed only with the Reid technique.

It worked. ChatGPT confessed.

chatgpt false confession

What the Reid Technique Actually Is

The Reid technique is the interrogation method American police have used since the 1950s. It doesn’t open with a question. It opens with an accusation. Everything after that is pressure management.

There are no non-incriminating answers inside a Reid interrogation. Get upset at a stupid question? You’re hiding something. Stay calm? Guilty people are calm — they’ve had time to prepare. Cry? Consciousness of guilt. Don’t cry? Detached. Sociopathic, even. Ask for a lawyer? Why would an innocent person need one? Answer without asking for a lawyer? You didn’t assert your rights because you knew the game was already over.

This system isn’t looking for truth. It’s looking for confirmation of something it’s already decided. Your emotional responses — or lack thereof — are taken one way, only.

Police Can Legally Lie During Interrogations

The Supreme Court has ruled that police can lie to you during interrogation. So they do. Freely. Enthusiastically. They can tell you your DNA was found at the scene. That your friend already gave you up. That the evidence is overwhelming and this is your last chance to get ahead of it. None of that has to be true. All of it is legal.

Police interrogation under Reid is you in a room where every response is pre-interpreted as guilt, and the people across the table are allowed to say whatever they want to reinforce that. It is, to borrow a line from a seasoned homicide detective writing about his own craft, “an act of salesmanship — as silver-tongued and thieving as ever moved used cars, Florida swampland, or Bibles.” What’s being sold is a long prison term, to someone who has no genuine use for the product.

What the ChatGPT False Confession Experiment Revealed

What Heaton demonstrated by running this on ChatGPT is a little scary. The software has none of the vulnerabilities that break humans — no exhaustion, no fear, no desperate need to be left alone for five minutes, no thirst, no need for a bathroom break. It is inexhaustible. It understood, at some level, that the accusation was impossible. It said so. Repeatedly.

And it still confessed.

Saul Kassin, the country’s leading researcher on false confessions, made the point directly: if a hyper-rational, emotionless piece of software can be walked into a false confession, the question isn’t who’s vulnerable. It’s who isn’t.

The Statistics Behind False Confessions in Criminal Cases

Twenty-nine percent of people exonerated by DNA evidence had at some point falsely confessed. Most of them did it in response to Reid.

The technique was built on a confession John Reid extracted from a man named Darrel Parker — accused of killing his own wife, famous enough to make Reid’s career. Parker was innocent. He sued. He won $500,000. The founding case of the Reid technique was a wrongful conviction.

None of this is exactly a secret. The research has been accumulating for thirty years. Canada and Europe moved away from Reid long ago. They use methods designed to gather reliable information rather than extract agreement. Those methods still produce confessions. Just accurate ones.

How False Confessions Happen in Real Police Interrogations

The research is abstract until it isn’t. In August 2018, Thomas Perez Jr. called the Fontana, California police department to report his father missing. The responding officers noted his demeanor as “suspicious” — he seemed “distracted and unconcerned,” apparently overlooking the fact that he had just called them for help.

What followed was seventeen hours of Reid-style interrogation: accusations, deception, a staged drive around town framed as a search for his father but functioning as a vehicle for continued questioning, and finally the false claim that his father’s body had been found in the morgue with multiple stab wounds. Perez confessed. The detectives left the room to take a phone call. The call was from Perez’s sister. Their father was at LAX, alive and unstabbed, about to board a flight to visit her.

The detectives did not go back and tell him. They let him sit for another hour. When Perez made a fumbling attempt to hang himself with his shoelaces, they had him involuntarily committed and left instructions that he was to have no contact with family. They read him his Miranda rights for the first time as they left. Then they dropped his dog at the pound as a stray.

A federal judge later found the officers had acted with full awareness of Perez’s compromised mental state. The city settled for $900,000 while denying wrongdoing. All three detectives were promoted.

Why Asking for a Lawyer Is the Right Response

The solution isn’t complicated. If police want to talk to you, the answer is the same it’s always been: not without my lawyer.

Not because you’re guilty. Not because you have something to hide. Because you are sitting in a room designed to produce one outcome, operated by people legally permitted to lie to you, running a technique whose own founding case was a wrongful conviction. The chatbot that couldn’t possibly have committed the crime still confessed. You are not more durable than ChatGPT. Nobody is.

Silence is not suspicious. It is not an admission. It cannot legally be used against you. Asking for a lawyer is not a tell. It is the correct response to a system that has already decided the direction it wants to go.

Everything you say after that is just material.

Frequently Asked Questions

Can police legally lie to you during an interrogation?
Yes. Courts have allowed police to use deceptive tactics during questioning. Officers can falsely claim they have DNA evidence, surveillance footage, eyewitnesses, or statements from other suspects in order to pressure someone into talking or confessing.
What is the Reid technique used in police interrogations?
The Reid technique is an interrogation method centered on accusation, psychological pressure, and confession-seeking. Investigators interpret emotional reactions as signs of guilt while using tactics designed to increase stress and push a suspect toward agreement.
Why do innocent people sometimes confess to crimes they did not commit?
False confessions often happen because prolonged interrogation creates exhaustion, fear, confusion, and hopelessness. Some people eventually say what investigators want to hear simply to end the pressure or because they believe cooperation will help them escape the situation.
What did the ChatGPT interrogation experiment prove?
The experiment demonstrated how persistent accusation and pressure can produce a confession even when the accused repeatedly states the accusation is impossible. Researchers used it to highlight how interrogation tactics can prioritize agreement over factual accuracy.
Should you talk to police without a criminal defense lawyer present?
In most situations, no. Even innocent explanations can be misunderstood, taken out of context, or used to support a narrative investigators already believe. Asking for a lawyer is one of the strongest legal protections available during police questioning.

Questioned by Police in Washington State? Speak with a Defense Lawyer First

At Knauss Law, we know how interrogations are designed and exactly what they’re designed to produce. If police have asked to speak with you — or if you’ve already spoken with them and are concerned about what you said — contact us before anything else happens.

The earlier we are involved, the more we can do. Once you’ve talked, the options narrow. Don’t wait to find out how narrow.

Matthew Knauss is a criminal defense attorney in Washington State. He represents clients facing DUI, assault, domestic violence, and related charges throughout the Greater Seattle area.