Anything You Ask ChatGPT May Be Used Against You in Court The Dark Side Of AI

Anything You Ask ChatGPT May Be Used Against You in Court-The Dark Side Of AI

By H. Michael Steinberg, A Colorado Criminal Defense Lawyer – Practicing Colorado Criminal Law from both sides of the courtroom for over 40 years.

Introduction:

A growing number of criminal cases across the country now share an unusual piece of evidence: a defendant’s own conversation with an AI chatbot.

---

• In Florida, a man charged with the murder of two college students had allegedly asked ChatGPT how a body might be discovered after being placed in a garbage bag.
• In Missouri, a college student was linked to a string of acts of vehicle vandalism after opening ChatGPT minutes after the incident and describing what he had done, then asking whether he would get caught.
• In Los Angeles, prosecutors cited a defendant’s chatbot history — including questions about fire liability — in an arson case tied to a major wildfire.
• In Virginia, a chatbot conversation about a hypothetical shooting became evidence in a first-degree murder prosecution.

None of these people were talking to a lawyer, a priest, or even a friend who might later be difficult for police to locate and question. They were talking to a chatbot — and in each case, investigators eventually obtained the conversation and used it in court.

There Is No Such Thing As a Privileged Chatbot Conversation

Many people understandably assume that a private conversation with an AI assistant carries some degree of confidentiality. It does not. Attorney-client privilege exists because society has decided that open, honest communication between a client and a lawyer serves justice — it allows people to receive accurate legal advice without fear that their words will be repeated in court.

No equivalent legal protection exists at this time for AI chatbots. There is no attorney-client relationship, no doctor-patient relationship, and no expectation of confidentiality that a court will recognize.

OpenAI’s own terms of service acknowledge that user conversations may be reviewed for safety purposes and disclosed in response to a subpoena or court order. Even messages a user deletes may remain on company servers for some time.

The same general framework applies across most consumer AI platforms. Once a conversation is created, it exists as a record — and records can be compelled.

If You Talk To…	Legal Protection	Risk of Disclosure
A Lawyer	Confidential. Protected by attorney-client privilege — cannot be forced into evidence.	None. The privilege is absolute for confidential communications seeking legal advice.
A Priest, Doctor, or Therapist	Confidential under recognized professional privileges in most circumstances – except mandatory reporting laws.	None, absent narrow statutory exceptions (e.g., imminent danger reporting duties).
An AI Chatbot	Not privileged. Stored on company servers and may be reviewed for safety or legal compliance.	Can be obtained by search warrant, grand jury subpoena, or Stored Communications Act order — and used against you.

How These Conversations End Up in a Criminal Case

Like other forms of social media, investigators do not need anything exotic to obtain an AI chat log. In practice, these records surface through the same tools used to obtain any other digital evidence:

• Search warrants. If police have probable cause — often from other evidence — a judge can authorize a warrant for a suspect’s phone, computer, or cloud accounts, which frequently includes an AI app with saved chat history.
• Consent searches. Here, a suspect voluntarily unlocks a phone for police, not realizing an AI chat history was sitting in the app.
• Subpoena to the AI company. Under the Stored Communications Act, prosecutors can subpoena identifying account information, and a warrant can compel the AI provider to produce the underlying content.
• Grand jury subpoena. In federal and some state investigations, a grand jury subpoena can reach both the AI provider’s records and any copies the defendant kept on his or her own devices.

 If You Are Under Investigation

If you believe you are being investigated for a crime in Colorado, the safest course is the same one criminal defense lawyers have given for decades — it simply now needs to be extended to a new category of “person” people talk to.

• Do not discuss the facts of your case with an AI chatbot. Treat it the same way you would treat a conversation with a police officer: anything you say can potentially be obtained and used against you.
• Do not use AI to “test” what might happen. Questions like “would I get caught” or “what happens if” read very differently to a jury than they do in the moment you typed them.
• Assume nothing is deleted. Chat histories, even those you believe you have removed, may still exist on a company’s servers or in a phone’s local cache.
• Talk to a lawyer, not a chatbot. A conversation with a licensed defense attorney is privileged. A conversation with an AI platform is not — no matter how much it feels like a private exchange.
• If you have already used a chatbot, say so — to your lawyer, not to police. Your attorney needs to know what exists so the defense can address it, and that conversation with your attorney remains protected.

“If you would not say it to a police officer, do not say it to a chatbot. The legal result, when it matters most, can be the same.”

A Closer Look: A Question Every Person Facing Criminal Charges Should Ask Before Using Artificial Intelligence

Artificial intelligence has become an extraordinarily convenient way to obtain information. A person can type a question into an AI system and, within seconds, receive an explanation of a criminal statute, an analysis of possible defenses, a discussion of sentencing, or suggestions about what might happen in court.

For someone facing a criminal investigation or prosecution, however, there is an important question that should come before the legal question:

“Is what I tell this AI system confidential and protected from disclosure to the police, prosecutors, courts, or anyone else?”

The safest answer is:

Do not assume that it is.

A conversation with an AI system is not protected by the attorney-client privilege merely because the subject is a criminal case or because the person is asking for legal advice. The privilege that protects confidential communications between a client and the client’s lawyer is fundamentally different from an ordinary conversation with a technology service.

That distinction can be extremely important in a criminal case.

AI Is Not A Criminal Defense Lawyer

The most important distinction is between using AI to obtain general legal information and communicating confidential information to your actual attorney.

When a person retains a criminal defense lawyer, communications made for the purpose of obtaining legal advice can, under applicable law, be protected by the attorney-client privilege.

The privilege exists because the law recognizes the importance of allowing clients to communicate candidly with their lawyers.

An AI system is not automatically an attorney, law firm, or protected legal representative.

Consequently, typing:

“What is the law concerning Colorado domestic violence assault?”

…is very different from typing:

“I punched my girlfriend in the face last Saturday at 10 p.m. in Denver, the police arrived, I told them I pushed her, and there is a surveillance camera showing what happened. What should I do?”

The first question is largely a request for general legal information.

The second potentially contains case-specific facts, admissions, identifying information, statements concerning conduct, and information that could be relevant to an actual criminal prosecution.

That is information a person should be extremely cautious about putting into an AI system.

A person might reasonably think:

“I’m having a private conversation with the AI, so the government can’t see it.”

That conclusion does not necessarily follow.

There are several different questions:

Does the AI company retain the conversation?
What does its privacy policy say about storage and use?
Who can access the information?
Can records be obtained through legal process?
Are there exceptions to ordinary privacy protections?
Would the information be subject to a subpoena or other court order?
Could the information become relevant in litigation?
Could the conversation be used as evidence if it were lawfully obtained?

Those are different questions from whether the conversation feels private to the person typing it.

A criminal defendant should therefore avoid treating an AI conversation as a privileged conversation with a defense attorney.

“Discoverable” Does Not Necessarily Mean “Admissible” At Trial

There is another important distinction.

People sometimes hear that an AI conversation could be “discovered” and conclude that the prosecutor could automatically introduce every word of it into evidence.

That is not necessarily true.

Discovery and admissibility are different concepts.

Discovery generally concerns information that parties may be entitled to obtain or inspect under the applicable rules and court procedures. Whether a particular item can ultimately be admitted into evidence is a separate question governed by rules concerning relevance, authentication, hearsay, privilege, constitutional protections, and other evidentiary principles.

Nevertheless, that distinction should not create false comfort.

Even if a particular AI conversation ultimately could not be admitted into evidence, its disclosure could still create serious problems.

For example, information contained in the conversation could:

– reveal facts the prosecution did not previously know;
– identify potential witnesses;
– reveal the existence of physical or electronic evidence;
– disclose inconsistencies in the defendant’s account;
– provide information investigators could use to locate additional       evidence;
– reveal potential defenses;
– disclose statements concerning the defendant’s conduct;
– provide material that could potentially be used for impeachment; or
– lead investigators to other evidence.

In other words, the danger is not limited to the question of whether the AI transcript itself can be shown to a jury.

The Biggest Mistake: Telling AI the Facts of Your Case

The greatest risk is usually not asking a general legal question. It is telling the AI what happened. Consider the difference.

Lower-risk question:

“What are the elements of second-degree assault under Colorado law?”

That is a general legal question.

Much higher-risk question:

“I hit the alleged victim once with my fist, but I only did it because she came toward me first. There were two witnesses, and I think one of them lied to the police. What defense do I have?”

Now the AI has been given potentially significant information about the actual case.

The person may have unintentionally created a written record containing statements that could become important if the information were later obtained.

The same principle applies to uploading documents.

A person should think carefully before uploading:

– police reports;
– body-camera transcripts;
– witness statements;
-photographs;
– text messages;
– emails;
– medical records;
– court documents containing sensitive information;
– discovery materials;
– probation records;
– recordings;
– police interview transcripts; or
– communications with a defense lawyer.

Those materials may contain information that should be kept within the attorney-client relationship and the defense team’s protected systems.

What About Asking AI After You Have Hired a Lawyer?

Having a lawyer does not automatically make your conversations with an AI system privileged.

Suppose a defendant tells the lawyer: “Here is what happened.”That communication may be protected by the attorney-client privilege.

But if the defendant then copies the same confidential information into an AI system, the defendant should not assume that the information retains the same level of protection simply because the original information was communicated to the lawyer.

This is one reason lawyers using AI must carefully consider how client information is entered into AI systems. The ABA has specifically identified confidentiality and data security concerns associated with the use of client information in generative AI tools. (American Bar Association).

What Should You Do If You Already Told AI Everything?

Don’t panic.

The fact that someone entered information into an AI system does not necessarily mean that the information will be obtained by law enforcement, admitted into evidence, or used against the person.

But if the information concerns an actual criminal investigation or prosecution, it is worth telling your defense attorney that you used an AI system and explaining generally what you disclosed.

Do not try to solve the problem by deleting evidence, destroying records, or attempting to conceal information from your lawyer or the court.

Instead, let your attorney evaluate the situation.

Your lawyer can determine whether anything about the AI conversation matters legally, whether any potential privilege or confidentiality issue exists, and what steps—if any—should be taken.

The Five-Second Rule Before Asking AI About a Criminal Case

Before entering information into an AI system, ask yourself five questions:

1. Is this about an actual criminal investigation or prosecution?

If yes, exercise caution.

2. Am I about to tell the AI what I actually did?

If yes, stop and reconsider.

3. Am I providing information that isn’t publicly known?

If yes, don’t assume it is protected.

4. Am I uploading a police report, discovery, recording, photograph, text message, or attorney communication?

If yes, consult your lawyer before doing so.

5. Would I be comfortable explaining this exact AI conversation to my defense attorney?

If the answer is no, don’t put it into the AI system.

Summary and Conclusion: Ask About the Law, Not About Your Crime

A useful practical rule is:

Use AI to learn about the law. Use your criminal defense lawyer to discuss the facts of your case.

For example, it is generally much safer to ask:

“What does probable cause mean in a Colorado criminal case?”

than:

“The officer arrested me after I admitted that I had the drugs in my backpack. Was there probable cause?”

It is generally safer to ask:

“What factors can affect whether a Colorado assault charge is a misdemeanor or felony?”

than:

“I broke the victim’s nose, but I didn’t mean to. What charge am I likely to get?”

And it is safer to ask:

“What is the general purpose of a preliminary hearing?”

than:

“The police report says X, but the body camera shows Y. How can I use that inconsistency at my preliminary hearing?”

The difference is simple:

The first category seeks legal education.

The second category begins revealing the facts and strategy of an actual criminal case.

The Bottom Line

Artificial intelligence can be an excellent educational resource, including for people trying to understand the criminal justice system.

But AI should not automatically be treated as a confidential substitute for a criminal defense lawyer.

A person facing criminal charges should assume that a detailed AI conversation about the facts of the case may not have the same confidentiality or privilege protections as communications with a defense attorney.

That does not mean every AI conversation is automatically discoverable, admissible, or available to prosecutors. Those questions depend on the particular circumstances, applicable law, procedural rules, and the manner in which the information was obtained.

But it does mean that a defendant should think carefully before creating a digital record containing admissions, confidential facts, defense theories, witness information, or other sensitive information.

The safest general principle is straightforward:]

And perhaps the most important question to ask before typing anything into an AI system is:

AI chatbots are remarkably good at feeling like a private, judgment-free place to think out loud. Legally, they are neither private nor judgment-free — they are a data record that police, prosecutors, and grand juries are increasingly learning to request. As these tools become part of daily life, expect Colorado prosecutors, like their counterparts nationwide, to routinely ask about AI usage during investigations and to subpoena chat histories as a matter of course.

If you are facing a criminal investigation or charge in the Denver metro area — whether or not an AI chatbot is involved — the most important call you can make is to a criminal defense lawyer before you make any other calls, searches, or chats.

---

Colorado Criminal Law – Anything You Ask ChatGPT May Be Used Against You in Court-The Dark Side Of AI

The reader is alerted that Colorado criminal law, like criminal law in every state and at the Federal level, changes constantly. The article appearing above was accurate when it was drafted, but it cannot account for changes occurring after it was uploaded.

ABOUT THE AUTHOR: H. Michael Steinberg – Email the Author at: hmsteinberg@hotmail.com

A Denver, Colorado Criminal Defense Lawyer – or call his office at 303-627-7777 during business hours – or call his cell if you cannot wait and need his immediate assistance – please call 720-220-2277.

“A good criminal defense lawyer is someone who devotes themselves to their client’s case from beginning to end, always realizing that this case is the most important thing in that client’s life.”

Putting more than 40 years of Colorado criminal defense experience to work for you.

You should be careful to make a responsible choice when selecting a Colorado criminal defense lawyer. We encourage you to “vet” our firm. Over the last 40 years – by focusing ONLY on Colorado criminal law – H. Michael has had the necessary time to commit to the task of constantly updating himself on nearly every area of criminal law, including Colorado criminal law and procedure and trial and courtroom practice.

H. Michael works hard to get his clients the best possible results in and out of the courtroom. He has written, and continues to write, extensively on Colorado criminal law, and he hopes this article helps you in some small way.