Your AI Chat Is Not a Private Journal: What the ChatGPT Case Says About Florida Law

A South Palm Beach man spent months telling ChatGPT what he wanted to do to his former girlfriend. He was not writing to her. He was not posting it where her friends would see it. He was typing into a chat window, in what must have felt like a closed room. In May a review team at OpenAI read those conversations and called the FBI. By August he had pleaded guilty to three felonies in Palm Beach County.

The story went national this week, and the headlines have carried the shocking part. What deserves a Florida lawyer’s attention sits underneath it, because the charge that carried this case only reaches a private conversation with a machine at all because of a change the Legislature made in 2023.

What the record actually shows

OpenAI’s own safety review flagged a pattern in the conversations, and the company alerted the FBI in May 2026. Federal agents passed roughly two months of chat logs to the Palm Beach County Sheriff’s Office. The man was arrested in May, spent two days in the county jail, and posted a $100,000 bond.

In June an assistant state attorney charged him with three felonies: aggravated stalking, written threats to kill or do bodily harm, and unlawful use of a two-way communications device. He pleaded guilty to all three on August 13, 2026. The circuit judge withheld adjudication and imposed eight years of probation with two years of electronic monitoring, along with counseling. The judge said on the record that he accepted the negotiated disposition because the woman involved had approved it.

This is a Florida case, and that changes everything about it

The FBI took the report and handed the material to a local sheriff’s office. Everything after that ran through the state attorney’s office and a Florida circuit court, which is easy to lose in a story about a federal agency getting a phone call from a technology company. It decides which statutes apply, which discovery rules apply, which evidence code applies, and which judge hears the motion. This was a Florida prosecution under Florida law from the day it was filed.

The disposition is a Florida creature too. Eight years of probation, two years of electronic monitoring, and a withhold of adjudication. On a withhold the court accepts the plea and imposes a sentence without entering a conviction, so finishing the probation leaves no felony conviction on the record. Violating it puts the whole thing back in front of the judge, who can then adjudicate and sentence up to the statutory maximum, which across these three counts reaches twenty five years. That is a long leash and a short one at the same time, and it is the trade at the center of most negotiated pleas in this state.

The statute, and the sentence that changed in 2023

The written threats charge comes from section 836.10, Florida Statutes. Subsection (2) reads, in full:

It is unlawful for any person to send, post, or transmit, or procure the sending, posting, or transmission of, a writing or other record, including an electronic record, in any manner in which it may be viewed by another person, when in such writing or record the person makes a threat to: (a) Kill or to do bodily harm to another person; or (b) Conduct a mass shooting or an act of terrorism.

A violation is a second degree felony. Compare the version that was on the books before the 2023 rewrite, as the First District quoted it in 2013:

Any person who writes or composes and also sends or procures the sending of any letter, inscribed communication, or electronic communication, whether such letter or communication be signed or anonymous, to any person, containing a threat to kill or to do bodily injury to the person to whom such letter or communication is sent … commits a felony of the second degree.

The old statute was built around a recipient. Courts read it to require three things: that a person composed a threat, that the person sent it or procured its sending to another person, and that the threat was aimed at the recipient or a member of the recipient’s family. The current statute drops the recipient. What it asks now is whether the writing was sent, posted, or transmitted in any manner in which it may be viewed by another person.

That is a much larger net, and a private-feeling conversation with a chatbot swims right into it, because the provider can view it and in this instance did.

The line a Florida court drew in 2013, and what dissolved it

The case worth reading beside this one is O’Leary v. State, 109 So. 3d 874 (Fla. 1st DCA 2013). A man posted a threat about a relative and her partner on his own Facebook page. He argued he had not sent anything to anyone, because he never addressed the post to a person or asked anybody to read it. The court disagreed, holding that by putting the composition somewhere his Facebook friends could see it, he had sent it.

Along the way the court drew the line that matters here:

Had appellant desired to put his thoughts into writing for his own personal contemplation, he could simply have recorded them in a private journal, diary, or any other medium that is not accessible by other people.

In 2013 that sentence described something obvious. There were places you published and places you did not, and everyone knew which was which. A diary in a drawer was safe because no one could read it.

A conversation with an AI assistant feels exactly like that diary. It is private in tone, it answers only you, and nobody else is in the room. It is also a hosted service whose operator reviews conversations, publishes policies saying so, and refers threats of serious harm to law enforcement. It is a medium that is accessible by other people, which is precisely the category the 2013 court used to describe the writing that gets you charged. The feeling and the legal reality point in opposite directions, and the statute follows the reality.

What lets a company hand your conversations to the government

The Stored Communications Act generally forbids a provider from divulging the contents of a user’s communications. The exception used here is 18 U.S.C. 2702(b)(8), which permits disclosure to a governmental entity if the provider, in the statute’s words, in good faith, believes that an emergency involving danger of death or serious physical injury to any person requires disclosure without delay of communications relating to the emergency.

Notice what that provision does not require. No warrant. No subpoena. No judge. The provider forms the good faith belief, and the provider decides what to send. Florida’s own statute points the same direction from the other side: section 836.10(3) exempts interactive computer services and other providers from liability for transmitting or storing someone else’s messages. The company that reports you is shielded, and the company that stays quiet is shielded too. Which way a provider goes is a matter of its policies rather than a matter of law.

The questions that never got answered

He pleaded, so none of this was litigated. That leaves the interesting questions open for whoever gets the next one, and there will be a next one:

• Is a conversation with an assistant a record transmitted in a manner in which it may be viewed by another person when the only viewer is the provider’s own review process? The text seems to reach it. No Florida court has said so.
• What carried the aggravated stalking count? A credible threat under section 784.048(1)(c) has to place its target in reasonable fear for her safety. If a message was never seen by the person it described, the fear has to come from somewhere else in the record, which means the chats and the course of conduct are doing different work and should be examined separately.
• What did the good faith belief in an emergency rest on, when was it formed, and what was reviewed before it was formed?
• How complete are the logs? An agency received about two months of conversation selected by someone outside the case. The selection criteria, the material left out, and the authentication of what came in are all fair ground.

None of that is a defense to what those messages said. It is the ordinary work of testing the State’s proof, which is what the process is for even when the underlying conduct is ugly.

The Mental State the State Has to Prove, and the Florida Courts Still Have Not Settled

Everything above concerns what the statute reaches. This section concerns the harder question, which is what the State has to prove was going on in the writer’s head. That question is currently unsettled in Florida, and the reason it is unsettled is worth knowing.

Start with the federal floor. In Counterman v. Colorado, 600 U.S. 66 (2023), the Supreme Court held that a true threats prosecution requires proof of the speaker’s subjective mental state. It is not enough that a reasonable person would have read the words as threatening. The Court set recklessness as the constitutional minimum, and defined it plainly: in the threats context, it means “a speaker is aware that others could regard his statements as threatening violence and delivers them anyway.”

The Court explained why the line has to sit somewhere: the word “true” in “true threat” is what “distinguishes what is at issue from jests, ‘hyperbole,’ or other statements that when taken in context do not convey a real possibility that violence will follow.” Without a subjective element, prosecutions would reach people who never understood their words that way, and the risk of that would quiet a great deal of lawful speech.

Now Florida, where the picture is genuinely muddled. The district courts of appeal have not agreed on whether section 836.10 carries a mental state element at all. The Fifth District held in 2003 that the statute did not require actual intent to do harm or any ability to carry the threat out. The Second, Third, and Fourth Districts have gone the other way. The Fourth District’s 2023 decision explains the split as a question of statutory vintage: the older cases construed an earlier version of the statute that lacked an intent requirement, and those courts measured a threat by its effect on the person who received it. Reading the current text, that court concluded a mental state element has to be read into the statute, and described it as requiring that the defendant made the communication knowing it would be viewed as a threat.

Notice that Florida’s formulation and the Supreme Court’s are not the same word. Knowledge is a higher bar than recklessness. Nobody has squared the two.

Why it stays unsettled

In March 2026 the First District affirmed a conviction under this statute in Dominguez v. State, No. 1D2024-1250. The defendant had posted a video holding a real semi-automatic rifle and naming his former middle school, with an on-screen disclaimer saying the video was a work of fiction. He received three years in prison followed by ten years of probation.

The court had the chance to decide what mental state section 836.10 requires. It declined, and said why: there was no argument in the trial court that the recklessness standard from Counterman should apply, and on appeal the defendant offered no position on which standard governed. The issue was not preserved, so the court did not reach it.

That had already happened once. In 2024 the Sixth District likewise declined, on preservation grounds, to decide whether the statute requires specific or general intent.

So two Florida appellate courts in two years have had the constitutional question in front of them and passed, both times because it was not raised below. The law stays unsettled not because it is hard, but because the argument keeps arriving too late.

What that means in a case

If a written threats charge is on the table, the mental state argument belongs in the trial court, named, and in writing. Counterman should be cited by name, and the standard instruction should be examined against it, because Dominguez also holds that intent is ordinarily a jury question rather than a basis for a judgment of acquittal. That makes this a fight about the jury instruction more than a fight about dismissal, and the charge conference is where it happens. An argument raised for the first time on appeal is worth nothing.

Two other things from that decision are worth carrying. A disclaimer does not defeat intent as a matter of law, and a jury is free to read one as evidence that the writer expected to need a defense. In Dominguez, the defendant’s own published advice about using disclaimers so that fiction would not “become evidence” was treated as cutting against him. Labeling something as art does not decide the question, and in that case it made things worse.

What to take from it

If you use these tools, treat the conversation as a record. It is stored, it is reviewable by the company, it can be produced, and in Florida the written threats statute no longer cares whether you addressed it to anybody. The old mental model, where something counts only if you sent it to a person, stopped matching the statute in 2023.

If you defend these cases, add AI use to the intake conversation the way you already ask about texts and social media. Ask what the client typed, on which services, and over what period, because the logs may already be in the State’s file before anyone has asked the client a single question.

I was asked this in March

Five months before any of this, reporters called about a federal ruling that let a defendant’s AI chatbot searches into evidence. I told them the same thing then that the Florida case demonstrates now: you are not talking to a lawyer, so there is no privilege, and if you would not shout it out in public, be careful what you put in there. I spent a good part of March answering that question. The longest of those interviews is on Gulf Coast News Now, and the same commentary ran on AZFamily, WVNS, and CBS6 Albany. The video below covers the same ground.

Watch

Related reading: digital forensics, searches of digital devices and phones, stalking and cyberstalking injunctions, and criminal defense in Florida.

Further Reading

1. OpenAI, Helping people when they need it most, the company’s own statement that conversations flagged for threats to others are human-reviewed and may be referred to law enforcement, official policy post
2. Fla. Stat. 836.10, Written or electronic threats, full current text, Online Sunshine
3. O’Leary v. State, 109 So. 3d 874 (Fla. 1st DCA 2013), the private-journal line, full opinion, FindLaw
4. Yahoo News, Florida man tells ChatGPT he plans to kill his ex, OpenAI alerts FBI, August 2026
5. Futurism, reporting on the backlash to OpenAI’s law enforcement disclosures, 2026

This page is general information about Florida law, not legal advice, and it does not create an attorney-client relationship. It describes a publicly reported case from court proceedings and news accounts. Every case turns on its own record, and past results do not guarantee a similar outcome.