TL;DR: A 2026 federal ruling confirmed that conversations with a public AI chatbot are not protected by attorney-client privilege or work product. In a divorce, that is uniquely dangerous: the exact things people paste into ChatGPT (financial facts, custody strategy, messages from their lawyer) are exactly what opposing counsel can demand in discovery. Sharing privileged material with a public tool can itself waive the privilege, and a paid tier does not fix it if the terms let the provider review, train on, or disclose your data. AI is not the problem. Using a consumer chatbot outside the privilege is. The fix is to use AI inside the attorney-client relationship, on a platform that does not train on or disclose what you put in.
Open any divorce forum and you will find the same quiet confession: people are running their case through ChatGPT. They paste in bank statements to find hidden money. They paste in their spouse's texts to build a custody timeline. They paste in the email their own lawyer just sent and ask the bot to explain it. It is cheap, it is fast, and it feels private. It is not private. And in a divorce, that gap is not academic. It is the difference between a strategy your lawyer controls and an exhibit your spouse's lawyer reads aloud in court.
This is the question almost nobody asks before they hit enter: is it safe to tell ChatGPT about my divorce? Below is the honest answer, what a recent court ruling actually says, and how lawyers can use AI on a divorce matter without giving away the one protection their client is paying for.
What the 2026 privilege ruling actually said
In a 2026 ruling from the Southern District of New York, a federal court addressed, for the first time squarely, whether conversations with a public AI chatbot are protected by attorney-client privilege or the work-product doctrine. The answer was no, on three grounds. The AI is not your lawyer, so no attorney-client relationship exists. There is no reasonable expectation of confidentiality, because the provider's terms let it review inputs, train on them, and disclose them to third parties. And the chat was not a communication made to obtain legal advice from counsel. We unpacked the decision in detail in our piece on why your AI conversations are not privileged.
Sam Altman said the quiet part out loud back in 2025: there is no legal confidentiality when people use ChatGPT for sensitive matters, and in litigation the company could be compelled to produce those conversations. The court has now started to agree.
Why divorce is the worst place to test this
Most legal matters do not put your most sensitive facts directly into the hands of an adversary with a subpoena. Divorce does. Family law runs on mandatory financial disclosure and broad discovery, and the opposing party is, by definition, someone who used to know everything about your life and now has a lawyer. Anything that is not privileged is potentially discoverable, and once a court agrees your AI chats are not privileged, they become just another category of document your spouse's attorney can demand.
The cruel irony is that the most valuable-feeling uses are the most dangerous ones. The table below maps what people actually do in a chatbot to how it can resurface in their own case.
| What you paste into ChatGPT | How it can resurface in your divorce |
|---|---|
| Your spouse's text messages, to find a pattern | Produced in discovery; you disclosed your theory of the case |
| Bank statements, to spot hidden assets | A roadmap of what you were looking for and what you found |
| The email your lawyer just sent you | Arguably waives privilege on that communication |
| Draft custody arguments and your weak points | Your own admissions, in writing, with a timestamp |
| A recording or summary of a meeting with your attorney | The privileged conversation, now shared with a third party |
| Prep notes before a deposition | A preview of your testimony for opposing counsel |
Expect opposing counsel to start asking the obvious questions: did you use any AI tools to prepare, and produce all prompts, inputs, and outputs related to this matter. The defensive law-firm blogs telling clients to stop using ChatGPT are not being alarmist. They are reading the same ruling.
The waiver trap: privilege can be gone the moment you hit enter
Here is the part people miss. Privilege is not a permanent label that travels with information wherever it goes. It is a status that depends on keeping a communication confidential. The moment you share privileged content with a public AI tool, that act of sharing can itself be a waiver, the same way forwarding a confidential email to a stranger waives it. The privilege may already be gone by the time the response loads, and you cannot claw it back by later showing the output to your lawyer.
A paid plan does not make it private
The most common rebuttal is: I pay for ChatGPT Plus, or Claude, so I am protected. The court's reasoning applies to any platform, free, paid, or commercially licensed, if its terms reserve the right to review, train on, or disclose user data. On consumer tiers, the default is often that the provider may retain prompts and use them to train models unless you turn that off, and opting out of training does not erase the provider's right to retain or to disclose in response to legal process. The question is not free versus paid. It is whether the tool will give you a contractual confidentiality guarantee, and whether you are using it under a lawyer's direction as part of the representation, or outside it.
The escape hatch, and why it is HAQQ's whole point
The same ruling hinted at the answer. The outcome might have been different if a lawyer had directed the use of AI within a workflow designed to protect privileged communications. Read that carefully: privilege can survive when AI is used under attorney direction, as part of the legal services the lawyer is actually rendering, on a platform that does not train on or disclose what goes in. That is not a loophole. It is the entire design brief for a legal-grade AI tool, and the precise line between a consumer chatbot and a platform built for law.
What family lawyers should actually do: keep privileged material under your control
The takeaway is not stop using AI. Used correctly, AI is a force multiplier across the whole arc of a divorce matter. The difference is that it runs under your direction, on infrastructure you can give a confidentiality guarantee on. Here is the eight-phase playbook we see working.
| Phase | What AI does under attorney direction |
|---|---|
| Intake and onboarding | Turn interviews and document piles into a structured fact sheet and first timeline |
| Financial disclosure | Draft the financial affidavit, build the marital balance sheet, flag anomalies in statements |
| Children and support | Walk the support computation step by step; draft the parenting plan; map best-interest factors |
| Property and QDRO | Model division scenarios; scaffold the QDRO for a specialist to finish |
| Discovery and review | Draft requests; summarize large record sets with citations; find contradictions across dates |
| Drafting | First-draft petitions, motions, and the settlement agreement, every term flagged for review |
| Mediation prep | Build the settlement proposal and an issue-by-issue BATNA map |
| Court prep | Assemble the chronology, exhibit index, and hearing outline tied to the governing law |
Two rules make all of it safe. First, the tool must operate inside the attorney-client relationship, directed by the lawyer, as part of the representation. Second, it must not train on or disclose the inputs, by contract and by construction. Miss either and you are back to the consumer-chatbot problem, with a malpractice exposure on top, because a wrong support number or a fabricated citation is your error, not the model's.
If you are representing yourself
Pro-se litigants are the ones most exposed, because there is no attorney in the loop to direct the use and no engagement to anchor the privilege. If you are using AI to navigate your own divorce, assume anything you type into a general chatbot is potentially discoverable, and keep the genuinely sensitive material, your strategy, your weak points, anything your lawyer told you, out of it. Use AI to understand the process and the vocabulary, not to store the case. If you can afford even an hour of a lawyer's time, the highest-value way to spend it is having them direct how you use these tools.
HAQQ's take
We built HAQQ with the court's warning in mind. HAQQ does not use inputs to train its AI models. It is designed to keep privileged material under the customer's control, with an audit trail of what the AI did and what a human approved, and whether a communication stays privileged depends on the jurisdiction and the customer's legal arrangements. Its controls are aligned with SOC 2 and ISO 27001, and it is grounded against real legal sources rather than free-associating, because in family law a confident wrong answer about a support guideline is worse than no answer. The honest framing is simple: the question was never AI versus no AI. It is a tool designed to keep client material under your control, versus a free product where, when the tool is free, your client's data is the product.
Vyzkoušejte HAQQ AI zdarma
Zažijte sepisování a rešerše s podporou AI
Key Takeaways
- A 2026 federal ruling held that conversations with a public AI chatbot are not privileged and not work product.
- Divorce is the worst place to test this: mandatory disclosure and discovery mean non-privileged material can land in your spouse's hands.
- Sharing privileged content with a public AI tool can itself waive privilege; the protection may be gone the moment you hit enter.
- A paid tier does not fix it if the terms let the provider review, train on, or disclose your data.
- The court hinted privilege can survive when a lawyer directs the use of AI on a platform that does not train on or disclose inputs.
- Lawyers can safely use AI across all eight phases of a divorce matter, as long as it runs under their direction and privileged material stays under their control.
