Your ChatGPT chats can be used against you in court

Many people turn to ChatGPT in the middle of the night to vent their deepest thoughts, treating the cursor like an unbreakable digital vault.

That illusion is falling apart fast in courtrooms across the country.

According to a Washington Post review, chatbot conversations have surfaced as evidence in at least 12 public criminal and civil court cases over the past two years. From post-crime panic checks to shady corporate cleanup, what users typed into an AI prompt is now being read aloud to judges.

How your private prompts leak into legal discovery

The fundamental issue is simple: AI chatbots have no law degree and no medical license. In the eyes of the legal system, conversations with them enjoy none of the confidentiality privileges that protect chats with your attorney, doctor, or therapist.

Investigators often do not even need to drag OpenAI into court to get their hands on the transcripts. Frequently, police simply find them right on the suspect's phone.

Michael Price, litigation director of the Fourth Amendment Center at the National Association of Criminal Defense Lawyers, noted that many people willingly hand over their devices during police stops, even though they usually have the legal right to refuse without a warrant.

That was precisely the downfall of Ryan Schaefer, a Missouri State University student accused of smashing 17 cars in a campus parking lot. After agreeing to let officers search his phone, police found messages he sent to ChatGPT right after the vandalism: "How f----d am i bro... What if i smahsed the s--- oitta mutlipls cars... Is there any way they could know it was me." He pleaded guilty to felony property damage and received five years of probation in July.

In civil litigation, the discovery process is pulling AI logs into the open just as easily. In Michigan, a tire salesman accused by his former employer of stealing clients asked ChatGPT whether deleted Yahoo emails could still be recovered through a subpoena. A judge deemed the query evidence of trying to hide tracks and ordered him to pay legal fees.

Sometimes, the AI platform itself calls the authorities. In Florida, OpenAI software flagged a user, Darren Zhou, who was repeatedly outlining plans to harm his ex-girlfriend. Human reviewers determined there was an imminent threat, contacted the FBI, and the bureau tipped off local police. Zhou pleaded guilty this month to stalking and threats, receiving eight years of probation.

OpenAI's transparency reports reflect this accelerating trend: in the second half of 2025 alone, the company handed over account information in response to law enforcement requests for more than 80 accounts, more than four times the figure from a year earlier.

Claude is not your defense attorney

Some defendants have tried claiming privilege after the fact, only to be laughed out of court.

Financial executive Bradley Heppner argued that prosecutors should not see his prompts to Anthropic's Claude, where he explored potential defenses against fraud accusations. A New York federal judge flatly rejected the claim, noting that Claude is not an attorney and Heppner was not instructed by counsel to consult it. Heppner was convicted of securities and wire fraud in May.

Mike Morgan, a partner at Morgan & Morgan, pointed out that teenagers and everyday users routinely share things with bots that they would never tell their parents or therapists, completely unaware that an expert witness could dissect those prompts in a deposition.

Why it matters

For regular users, the takeaway is stark: never treat a cloud-based chatbot like a secret diary or confidential counsel. If you are dealing with a workplace dispute, drafting a divorce strategy, or navigating legal gray areas, typing those scenarios into ChatGPT effectively creates a subpoena-ready paper trail.

We think this exposes a massive paradox in the AI industry. Companies like OpenAI desperately want their tools to become intimate, ubiquitous life companions, yet their legal obligations require active surveillance and compliance. OpenAI will, and should, report credible threats of violence to the FBI, but that reality destroys any lingering marketing fantasy of absolute personal privacy.

As George Washington University law professor Andrew Ferguson observed, as AI handles more daily tasks, "your entire world is going to now be available for police." Until local, on-device models with true zero-knowledge encryption become the norm, users must assume that anything told to an AI can, and will, be used against them in a court of law.