You typed the messy version of your divorce into ChatGPT at 1 a.m. because it felt safer than telling anyone. Here is the verdict: it was not safe. Your ChatGPT, Claude, and Gemini conversations carry no privilege, they are discoverable, and a subpoena can reach them in a US divorce. Opposing counsel can serve a discovery request for all of your communications with AI tools, including prompts and outputs, and a family court can order you to produce them. That covers anything you typed about hiding assets, custody strategy, or your spouse, which a court can read as a record of your intentions.
A chatbot is not your lawyer and it is not your spouse, so neither attorney-client privilege nor spousal privilege attaches to what you type into it. Courts treat those messages as ordinary electronically stored information, which is exactly what the rules of discovery are built to reach. If you vented, planned, or drafted anything about the divorce inside an AI tool, assume the other side can ask for it.
This article is general information, not legal advice. Discovery rules, privilege law, and preservation duties vary by state and by judge. Talk to a licensed family-law attorney in your jurisdiction before acting on anything here.
Why AI chats have no privilege in a divorce
Privilege is a narrow legal shield, and a chatbot falls outside every category that would apply. Attorney-client privilege protects confidential communications between you and a lawyer made to get legal advice. A commercial AI model is not a lawyer and cannot form an attorney-client relationship, so the privilege never forms in the first place. Spousal privilege protects certain communications between married partners, not messages you send to a software product owned by a third party.
Confidentiality is the second problem. Privilege depends on a reasonable expectation that the communication stays private. Most consumer AI terms of service let the provider collect your inputs and outputs, use them to improve models, and disclose them to third parties in some situations. Sharing your planning with a service that reserves those rights undercuts any claim that you expected the conversation to stay secret.
In February 2026, on what the court called a question of first impression nationwide, Judge Jed Rakoff of the Southern District of New York held that a defendant's exchanges with a consumer AI tool carried no attorney-client privilege and no work-product protection. The case was United States v. Heppner, and the tool was the consumer version of Anthropic's Claude. Rakoff reasoned that Claude is not an attorney, which alone defeated the privilege, and that the provider's own policy letting it collect user inputs and disclose them to third parties, including government regulators, killed any reasonable expectation of confidentiality. The case was criminal, not a divorce, but family courts apply the same privilege logic.
What opposing counsel can actually request
Discovery in a divorce is broad. Under state civil procedure rules, a party can request any non-privileged material that is relevant to the case, and AI chat logs qualify. In New York, for example, the same framework that governs document discovery lets a spouse seek AI conversations covering finances, custody strategy, and settlement thinking.
A request often reads like a template: produce all communications with AI-based tools, including every prompt and every output, relating to the marriage, marital assets, income, the children, or the divorce. Expect opposing counsel to ask about your AI use in a deposition. If they suspect you are holding back, follow-up subpoenas come next. If the chats live in your account or on your devices, you are the custodian, and the duty to produce lands on you.
The provider can be reached too
Even setting aside your own account, courts have already ordered AI providers to hand over logs. In the consolidated copyright litigation against OpenAI in the Southern District of New York, Judge Sidney Stein affirmed an order in January 2026 compelling OpenAI to produce 20 million ChatGPT conversation logs. That set was a de-identified sample, the case is about copyright rather than divorce, and it named no single person's history. The court accepted that users keep some privacy interest, then found it outweighed, in part because users voluntarily submitted their conversations to the service. The lesson still lands: legal process can pull the logs, and no product setting locks them away.
The prompt that becomes evidence of intent
Some AI chats do more than reveal strategy. They document intent. Type how do I hide assets from my spouse, or how do I move money before filing, and you have created a written record of a plan a court will treat as bad faith. In divorce, hiding or dissipating marital assets can lead to sanctions, an unequal division that favors the other spouse, and a serious credibility hit that colors everything else the judge decides.
The same danger runs through custody. A prompt like how do I make my ex look unstable in court, or drafting a misleading timeline of events, can undercut your position if it surfaces in discovery. Judges weigh candor heavily in family court. A chat log showing you asked a machine to help you deceive the court or your co-parent is difficult to explain away.
There is a subtler trap too. Even honest venting can be twisted. A late-night message where you say you want to make your spouse pay, or that you would do anything to keep the kids, reads very differently in a courtroom than it felt at 2 a.m. Tone does not carry. A judge sees the words on the page, stripped of the frustration that produced them, and opposing counsel will frame them in the least flattering light possible. What you meant as a way to blow off steam can be recast as evidence of a plan or a temperament.
Before you type anything into an AI tool during a separation, ask whether you would be comfortable reading it aloud in front of the judge deciding your case. If the answer is no, do not type it there.
Why deleting the chats can make things worse
Here is what most divorce guides will not tell you: the delete button is the trap, not the escape. Deleting your AI history after you reasonably anticipate a divorce is one of the most damaging moves you can make. Once litigation is reasonably foreseeable, a duty to preserve relevant electronically stored information attaches, and AI chat logs qualify. Wiping them can be spoliation, and courts hold specific tools to punish it.
Federal Rule of Civil Procedure 37(e) is the reference point, and many states model their own rules on it. When ESI that should have been preserved is lost because a party failed to take reasonable steps, and it cannot be restored, a court may order measures no greater than necessary to cure the prejudice. If the court finds a party acted with the intent to deprive the other side of the information, it may go further: it can presume the lost data was unfavorable, instruct the jury or fact-finder to presume as much, or even dismiss the claim or enter a default judgment.
In plain terms: deleting the chat can create the exact adverse inference you were trying to avoid. A judge who learns you erased AI conversations right after your spouse filed may simply assume the deleted messages were bad for you and rule as if they were. In state family court the labels differ, and remedies can include adverse inferences, striking pleadings, or monetary sanctions, but the direction is the same.
The preservation duty can attach before anything is filed. If you have hired a lawyer, sent a demand letter, or told your spouse you want a divorce, litigation may already be reasonably anticipated, which means the clock on preserving evidence has started.
Your lawyer vs a chatbot: who can read it later
| Dimension | Talking to your divorce lawyer | Typing into a consumer AI chatbot |
|---|---|---|
| Privilege | Protected by attorney-client privilege when confidential and for legal advice | No privilege; the AI is not a lawyer and not your spouse |
| Confidentiality | Duty of confidentiality binds your attorney | Provider terms often allow collection, model training, and third-party disclosure |
| Discoverable by your spouse | Generally shielded from discovery | Prompts and outputs are discoverable ESI |
| Can be subpoenaed | Generally shielded; privilege can be asserted against a subpoena | Your account and the provider's logs can be reached by legal process |
| Effect of deleting it | No duty to delete; your lawyer preserves the file | Deleting after litigation is anticipated can be sanctionable spoliation |
Practical steps if you use AI during a divorce
You do not have to swear off AI. You do have to be deliberate about what it sees and where the record lives.
- Assume every prompt and output is discoverable. Write as if a judge and your spouse's attorney will read it.
- Keep legal strategy with your lawyer. Communications made to get legal advice from your actual attorney can stay privileged. A chatbot cannot give you that protection.
- Do not ask an AI how to hide assets, move money, or make your ex look bad. That is a written record of intent, and it can sink your case.
- Do not delete chat history once you anticipate the divorce. Preserve it and tell your attorney what exists so they can advise you on production.
- Read the provider's data and retention terms. Know whether your inputs are used for training and how long logs are kept.
- Keep genuinely private planning out of a provider-controlled chat store in the first place. What is never written into a discoverable system cannot be produced from it.
Where your private planning lives matters
The safest chat is the one that was never stored in a system built to be searched and produced. That is the practical takeaway from every case above, reviewed as of July 2026 by Aditya Kumar Jha, a founding engineer at MemX who works on private memory infrastructure. Privilege will not save a consumer AI conversation, and deleting one after the fact only deepens the hole.
This is the gap MemX is built to address. MemX is an external, user-owned memory layer for tools like ChatGPT, Claude, and Gemini, designed to be private by architecture: per-user isolation, encryption at rest, and a stated policy of not training on your data. The point is control over where your private context lives, so your planning does not accumulate inside a provider's discoverable chat store by default. Every conversation you keep out of a provider-owned log is one fewer document a subpoena can name.
To be clear about the limits: no memory layer, including MemX, is immune from a lawful subpoena or a valid court order. MemX is not end-to-end encrypted and it is not zero-knowledge. Nothing here promises that any data can be placed beyond the reach of a court. The goal is fewer copies of your private planning sitting in systems you do not control, not a claim of legal immunity.
Frequently asked questions
01Can my ChatGPT history be used against me in a divorce?
Yes, and there is no privilege to stop it. AI chats are electronically stored information. If they are relevant to finances, custody, or your intentions, opposing counsel can request them in discovery and a court can order you to produce them.
02Are conversations with ChatGPT protected by attorney-client privilege?
No. A chatbot is not a lawyer, so no attorney-client relationship or privilege forms. In United States v. Heppner (SDNY, February 2026), a court held that consumer AI exchanges were not privileged or protected as work product.
03Can my spouse subpoena my AI chats?
Yes. Your wife or husband can reach your own account because you control it, and courts have also ordered AI providers to produce logs through legal process. Spousal privilege does not apply. You are communicating with software, not your spouse.
04Is it illegal to delete my ChatGPT history before a divorce?
Deleting relevant chats after you reasonably anticipate a divorce can be spoliation of evidence. Under FRCP 37(e) and similar state rules, that can trigger an adverse inference, other sanctions, or a presumption that the deleted material was unfavorable to you.
05How do I use AI safely during a divorce?
Assume everything is discoverable, never ask how to hide assets or deceive the court, keep real legal strategy with your attorney, and do not delete history once litigation is anticipated. Keep truly private planning out of provider-controlled chat stores. The simplest test: if you would not read it aloud in front of the judge, do not type it there.
