Can my spouse really use my AI chats against me in a divorce?

Yes, and it has already happened. In G v K [2025] EWHC 2961 (Fam), a father produced screenshots from his wife's ChatGPT account — including questions about separating — in High Court proceedings. He had access to the account; no subpoena or court order was involved. Shared accounts, shared devices and known passwords are the most common route by a wide margin.

Are my AI conversations about money and debt confidential?

Not in a legal sense. AI chats carry no attorney-client, financial-adviser or other privilege, and in United States v. Heppner (S.D.N.Y., February 2026) a court held that protection never attached to a defendant's chatbot-generated documents at all. A handful of 2026 civil rulings have protected AI-assisted litigation preparation as work product, but that is a narrower and more losable doctrine, and it does not reach the conversation you had before you had a lawyer.

Could my debt questions affect the ads I see?

Potentially. OpenAI began testing ads in ChatGPT in February 2026 on the Free and Go plans, and with personalized ads enabled the signals can include past chats and memory. OpenAI states that ads are not eligible to appear near sensitive or regulated topics — it names personal health, mental health and politics. Personal finance is not on that list, and financial-services advertisers can be eligible to advertise. Advertisers do not receive your chats or memories, and personalization can be switched off.

Does deleting the conversation fix it?

Only against a casual reader. Deletion removes a chat from your view, but a preservation order can require a provider to retain copies regardless — that is what happened in the New York Times litigation against OpenAI — and it does nothing about screenshots, a shared account, or a memory feature that already saved the summary. Deleting material once you are a party to a dispute can also create its own, worse problem.

Does Secret Chat let me hide assets or evidence?

No, and you should not try. Secret Chat is an anonymizer for the lawful privacy of ordinary users: it keeps no server-side chat archive and does not link your queries to your identity. It does not create privilege, it is not immune to legal process, and it does not affect your duty of financial disclosure or your obligation to preserve relevant material. Responsibility for lawful use stays with you — see our Terms and Disclaimers.

Talking to AI About Money, Debt and Divorce

Published August 14, 2026

This article is general information, not legal or financial advice. Laws differ by country and are changing fast. For your own situation, talk to a qualified lawyer or adviser — not a chatbot. Your use of Secret Chat is governed by our Terms of Service and Disclaimers, which make clear that you are responsible for your own input and for using the service lawfully.

It is nearly always late, and it is nearly always the same three sentences. "I think my marriage is over. We have a joint mortgage and about eleven thousand on cards in my name. What happens to the debt if I leave?"

You would not say that to a colleague. You have not said it to your mother. You may not have said it out loud at all. But the chat window is right there, it is patient, it does not react, and it knows more about matrimonial finance than anyone you could call at that hour. So you type it.

In September 2025, a judge of the High Court in London read screenshots of exactly that kind of conversation. They had been put in front of him by the woman's husband.

The Case Where Nobody Needed a Subpoena

The case is G v K [2025] EWHC 2961 (Fam), a Hague Convention child-return case heard in the Family Division by HHJ P Hopkins KC and decided on 29 September 2025. Buried in the evidence is a detail that should be far better known than it is. The judge records it plainly:

"He had access to her ChatGPT account, which I confess was new to me. This was explained to me as effectively analogous to one person having remote access to another person's internet search history. I was taken to screen shots where the mother appears to ask for advice about separating and was indicating an 'affair partner'."

Read that again, because every word of it matters. There was no subpoena. There was no data breach, no forensic examiner, no court order to a provider, no hacking. A husband had the password to his wife's AI account, and her private questions about leaving him arrived in a High Court bundle as screenshots. The judge, to his credit, found the whole thing novel enough to say so.

It went further. Later in the judgment he notes that the father became aware the mother "searched on ChatGPT for evidence she would need to report domestic violence to the UK police" — and adds, drily, "I digress to note that it follows he was still seemingly accessing her account at that time."

Almost every article about AI and legal risk — including ours about subpoenas — starts with courts and providers, because that is where the dramatic rulings are. G v K is a useful corrective. The most likely reader of your divorce prompts is not a judge, a prosecutor or an engineer at a model provider. It is the person whose interests are opposite to yours and who already knows your password.

Why Money, Debt and Divorce Belong Together

These three subjects are usually filed apart — one is financial, one is emotional, one is legal. From a privacy point of view they are a single category, and they behave unlike anything else you type into an AI.

Somebody's interests are directly opposed to yours. Ask an AI about a rash and no one is on the other side. Ask it about splitting a pension, arrears on a car loan, or what your spouse is entitled to, and there is a specific person or institution who would benefit from reading the answer — and often from reading the question even more.
The prompt reads as intent, not curiosity. "How much would I get in a divorce" cannot be idle. "Can I stop paying this card without being sued" cannot be research. These sentences have no innocent second meaning, which is what makes them so useful to anyone quoting them back at you.
Nobody built you a carve-out. Health has one. Mental health has one. Politics has one. Money — as we are about to see — does not.

There is a fourth property that is easy to miss: this is the category where people deliberately go to a machine because there is no human they can ask. You cannot ask your partner what leaving them would cost. You may be too ashamed to tell a friend the real card balance. The AI gets the unedited version — the actual numbers, the actual dates, the actual name of the affair partner — precisely because nothing else in your life is safe enough to hear it. The candour that makes the answer useful is the same candour that makes the record devastating.

The One Kind of Distress With No Protected Status

In February 2026, OpenAI began testing ads in ChatGPT — a US test that started on 9 February and has expanded since. Ads may appear for users on the Free and Go plans; Plus, Pro, Business, Enterprise and Edu accounts do not have them. So far, so ordinary.

The interesting part is in the fine print of OpenAI's own help page on ads, and it is a genuinely careful document. Ads, it says, "are not eligible to appear near sensitive or regulated topics, including personal health, mental health, or politics." Advertisers "never receive your chats, chat history, memories, name, email, precise location, IP address, or sensitive information (such as health, mental health, political topics)."

Notice what is on that list twice, and what is on it not at all. Health is protected. Mental health is protected. Politics is protected. Debt is not. Insolvency is not. Divorce is not. A page later, the same document confirms that "some advertisers in sensitive or regulated verticals/categories, such as health and financial services, may be eligible to advertise if they meet our strict eligibility criteria."

And with Personalized Ads enabled, the signals that select what you see can include your current chat thread, how you interact with ads, and — the phrase that matters here — "past chats and memory."

Put the pieces together and the result is not sinister, but it is worth seeing clearly. A person in real financial trouble is, by a distance, one of the most commercially valuable audiences on the internet: consolidation loans, debt-management plans, equity release, claims firms, "we buy any house". That audience is exactly who types "I'm four months behind and they're threatening a charging order" into a chat box. The topic that most reliably signals financial desperation sits outside the sensitive-topic fence that health, mental health and politics sit inside.

To be fair to OpenAI, which is more transparent about this than most: advertisers do not get your conversation, personalized ads can be switched off, and paid tiers have no ads at all. This is not a story about your debts being sold. It is a story about which categories of human distress somebody decided to draw a line around — and about the fact that a distinctive, durable inference ("this user has a debt problem") can be derived from your chats whatever the ad rules permit today. Ad rules are settings. Settings change. The inference, once formed, is the thing that persists — a pattern we go through in detail in AI memory as a privacy time bomb.

A Search Is Not a Statement — but It Will Be Read as One

Return to G v K for the most underrated risk of all.

The mother's searches were, on the father's case, evidence of a plan. The judge declined to treat them as proof of one, and his finding is the sentence to remember: "the ChatGPT reference is open to interpretation in a number of ways, including the father's assertions and the actions of a genuine victim of abuse trying to gain an understanding of how to report abuse."

That is a careful judge reaching a fair result. But look at the cost of getting there. A woman's questions to a chatbot about how to report domestic violence became a live issue in a High Court case argued by King's Counsel on both sides, and a judge had to weigh what her own search had meant. The point did not go against her. She still had to have it decided.

This is what people get wrong about AI logs in a money or family dispute. The danger is rarely a confession. It is that a question — asked in fear, asked at 2am, asked in a state of mind you would not want quoted — is a blank canvas that the other side gets to paint on first. You asked "what happens if I move money out of the joint account" because you were frightened of being left with nothing. It reads, in a bundle, in twelve-point type, next to a date stamp, as planning. You will get to explain. Explaining is the cost.

What the Law Actually Does With These Records

The legal position deserves precision rather than alarm, and we cover it at length elsewhere. In short:

There is no privilege in a chatbot. In United States v. Heppner (S.D.N.Y., 10 February 2026), Judge Rakoff held that a defendant's Claude-generated defence documents were protected by neither attorney-client privilege nor work product — the protection never attached in the first place, rather than being waived. Privilege comes from a lawyer, not a tool. The full analysis is in our article on AI and attorney-client privilege.
Some civil courts have protected AI work product — narrowly. Several 2026 decisions have, among them Assini v Hayward (Sup. Ct. Nassau County, 4 June 2026), which quashed a subpoena served on OpenAI for a litigant's whole account. But work product is a different, more losable doctrine that turns on litigation being anticipated and on who prepared the material. It protects a party preparing a case. It says nothing about the chat you had eight months before you had a lawyer.
You do not have to be a party to be swept up. In the New York Times litigation, a court ordered OpenAI to produce a 20-million-conversation sample. By OpenAI's own account it was a random sample of consumer conversations from a fixed date range — inclusion turned on the conversation having been retained and falling in that dataset, not on who you are or what you discussed. Nobody in it was notified or got to object.
The delete button loses to a legal hold. Deleting a conversation removes it from your view. A preservation order can require the provider to keep copies anyway, as happened in that same case. See the fine print of temporary chats for what the privacy modes really do.

None of this is a reason to panic. It is a reason to be deliberate about which conversations you commit to a system that keeps records, and which you have somewhere that does not.

The Line This Article Will Not Cross

"Keeping it off the record" means one thing here, and it is worth stating flatly so that nobody misreads the rest.

It means not generating gratuitous, permanent, searchable records of lawful private thinking — the 2am fear, the half-formed plan you abandoned by breakfast, the number you were too ashamed to say out loud. That is ordinary privacy, and you are entitled to it.

It does not mean hiding assets, defeating financial disclosure, or destroying material a court is entitled to see. In a divorce, both parties owe a duty of full and frank financial disclosure, and once litigation is contemplated your duty to preserve relevant material stands regardless of which tool you used. If you are a party to a case, deleting things to keep them from the other side is a category of problem that no privacy setting solves and that routinely damages the person who does it more than the underlying facts would have. And if what you actually want is a way to move money where it cannot be found, the honest answer is that this is not that article, Secret Chat is not that product, and the right person to talk to is a lawyer — who will tell you the same thing.

Keeping It Off the Record: What Actually Works

1. Get off the shared account before you need to

This is the single highest-value action in this article, and it costs nothing. If you share an AI account, a family plan, a laptop, a browser profile or a phone that mirrors notifications, do not have the difficult conversation there. Not in a temporary chat, not signed into a second profile in the same browser, not "just this once". G v K happened because the account was reachable, and the moment to fix that is before the first prompt, not after — a sudden password change by someone who has just started researching separation is itself a signal, and the history is already there to be read. The same logic applies to a work account, where an employer administers the workspace and can be compelled to search it; we cover that in never job-hunt in your work AI account and what AI on a shared computer can see.

2. Ask the general question, not the identifying one

Almost all of the value you want is in the general answer. How matrimonial debt is usually treated, how a statute-barred debt works, what a consent order does, what a creditor can and cannot do — none of that needs your name, your bank, your spouse's employer, your children's school or the exact balance to the penny. Round the numbers. Drop the names. Use "my partner" rather than the name you would be quoted saying. You will get the same answer, and what remains in the record is a person who once asked a general legal question, which is not evidence of anything.

A prompt that makes this the default for the whole conversation:

I want to think through a personal financial and family situation. I will use round numbers and no real names, places or employers, and I would like you to answer generally rather than asking me for identifying details. If a detail genuinely changes the answer, tell me which one and why instead of asking me to supply it. Here is the situation:

3. Do not upload the documents

There is a large difference between describing a situation and uploading the bank statements, the mortgage offer, the payslips or the letter from the solicitor. A typed summary is a paraphrase. An uploaded file is the original, complete with account numbers, addresses, employer names, transaction histories and — for anything photographed or scanned — the metadata riding along inside it. If you must upload, crop and redact first. Note too that the moment you paste in your spouse's or your ex-partner's financial details, you have disclosed their personal data, and that is your responsibility, not the tool's.

4. Turn off the memory before the hard conversation, not after

Personalization features exist to carry facts forward. That is the entire point of them, and it works: a tool that has learned "this user is separating and has significant unsecured debt" can carry that forward into unrelated answers — including one that appears on a screen you are sharing in a meeting, or on a phone your partner picks up. Turn off memory, chat history training and ad personalization in your settings before the conversation you care about, because switching them off afterwards does not unlearn what was already saved. Our AI privacy settings checklist and opt-out guide have the exact steps per platform.

5. Keep the legally loaded parts with a human

Use AI to understand the landscape — the vocabulary, the process, the questions worth asking. Take the strategy to a lawyer. That is not a counsel of perfection; it is the only route to actual privilege, and it also produces better outcomes, because the questions that matter in a real financial settlement are usually the ones you did not know to ask. A chatbot is an excellent way to arrive at a first meeting already informed. It is a poor substitute for the meeting.

6. Assume the record outlives the feeling

The mood that produces the most damaging prompt lasts about twenty minutes. The record lasts as long as somebody's retention policy, a litigation hold, or a backup nobody remembered to prune. Before typing the sentence you would not want read aloud, it is worth asking the only question that reliably helps: if this appeared in a bundle in two years, next to a date, with no context, what would it look like?

How Secret Chat AI Fits — and Its Honest Limits

Secret Chat AI exists for exactly the kind of conversation described at the top of this article, and it is worth being precise about what that does and does not mean.

What it does. It builds no profile of you. No chat is ever associated with you. Your conversations live only in your own browser, and a prompt exists on our side only for as long as it takes to fetch your answer — there is no stored chat archive on our servers, so there is nothing here for an opposing party to demand from us and nothing to accumulate into an inference about your finances. Your queries reach the top LLMs anonymously, and they are never used for training. Registration takes an email, but it is used only for account access and payment — never stored against your prompts or associated with them. Whatever a model provider holds under its own terms and configuration, the record on that side carries our gateway's credentials and server address, not your name, your account or your IP. You use the model as a stranger. There is also no memory feature quietly building a dossier out of the worst month of your life, and no advertising system on any tier with an opinion about your debts.

What it does not do. Secret Chat AI removes you from your queries — it does not remove the data from your messages. "Anonymously" describes the link, not the words: no account identifier travels with your prompt, but that is not a claim that the text stops being identifying. Write your spouse's name, your account number or your case reference into a message and it is all still sitting there in the message, reaching the provider verbatim. Redacting identifying details before you send is your responsibility, here as anywhere — which is exactly what step 2 above is for. We keep ordinary infrastructure telemetry (IP addresses, request timestamps, error codes) for at most 30 days, as our privacy policy sets out; that is not a record of what you asked. And anonymity is not privilege, not a legal exemption, and not a licence to hide anything a court is entitled to: only a lawyer creates privilege, your duty to preserve and disclose relevant material is unaffected by the tool you use, and the copies in your own browser remain fully discoverable from you. The honest, narrow value is real — for the lawful private thinking of ordinary people at the worst moment of their lives, far less of it ends up in a database with their name on it.

Frequently Asked Questions

Can my spouse really use my AI chats against me in a divorce?
Yes, and it has already happened. In G v K [2025] EWHC 2961 (Fam), a father produced screenshots from his wife's ChatGPT account — including questions about separating — in High Court proceedings. He had access to the account; no subpoena or court order was involved. Shared accounts, shared devices and known passwords are the most common route by a wide margin.
Are my AI conversations about money and debt confidential?
Not in a legal sense. AI chats carry no attorney-client, financial-adviser or other privilege, and in United States v. Heppner (S.D.N.Y., February 2026) a court held that protection never attached to a defendant's chatbot-generated documents at all. A handful of 2026 civil rulings have protected AI-assisted litigation preparation as work product, but that is a narrower and more losable doctrine, and it does not reach the conversation you had before you had a lawyer.
Could my debt questions affect the ads I see?
Potentially. OpenAI began testing ads in ChatGPT in February 2026 on the Free and Go plans, and with personalized ads enabled the signals can include past chats and memory. OpenAI states that ads are not eligible to appear near sensitive or regulated topics — it names personal health, mental health and politics. Personal finance is not on that list, and financial-services advertisers can be eligible to advertise. Advertisers do not receive your chats or memories, and personalization can be switched off.
Does deleting the conversation fix it?
Only against a casual reader. Deletion removes a chat from your view, but a preservation order can require a provider to retain copies regardless — that is what happened in the New York Times litigation against OpenAI — and it does nothing about screenshots, a shared account, or a memory feature that already saved the summary. Deleting material once you are a party to a dispute can also create its own, worse problem.
Does Secret Chat let me hide assets or evidence?
No, and you should not try. Secret Chat is an anonymizer for the lawful privacy of ordinary users: it keeps no server-side chat archive and does not link your queries to your identity. It does not create privilege, it is not immune to legal process, and it does not affect your duty of financial disclosure or your obligation to preserve relevant material. Responsibility for lawful use stays with you — see our Terms and Disclaimers.

Conclusion

The reason people type these questions into an AI is not carelessness. It is that money, debt and the end of a marriage are the subjects with the fewest safe listeners — and a machine that answers at 2am without flinching is a genuinely useful thing to have. That instinct is sound. What deserves a second thought is only where the conversation happens.

A judge in London has now confessed, on the record, that reading a spouse's ChatGPT history was new to him. It will not be new to the next judge. Before then, the cheapest protection available is also the most old-fashioned: have the conversation somewhere that keeps no archive, on an account nobody else can open, in words you would not mind being read back to you. Say what you need to say to get a real answer — and leave out the details that only ever served to identify you.

Money, Debt and Divorce:
How to Get AI Help Without Leaving a Record