On April 16, 2026, family law attorneys across the United States issued a coordinated warning to divorce clients: conversations with ChatGPT, Claude, Gemini, and other consumer AI chatbots are not protected by attorney-client privilege. A February 17, 2026 federal ruling out of Manhattan ordered the production of 31 AI chatbot documents — the first decision of its kind — meaning New York spouses who type financial details, custody strategy, or settlement goals into AI tools can have every prompt subpoenaed during discovery.
Key Facts
| Item | Detail |
|---|---|
| What happened | Federal court compelled production of 31 AI chatbot conversations; national attorney warning issued |
| When | Ruling: February 17, 2026; coordinated warning: April 16, 2026 |
| Where | Southern District of New York (Manhattan); guidance extends nationwide |
| Who is affected | Any divorce litigant using ChatGPT, Claude, Gemini, Copilot, or Perplexity for case-related prompts |
| Governing rule | CPLR § 4503 (attorney-client privilege); CPLR § 3101 (discovery scope); Fed. R. Evid. 502 |
| Practical impact | AI prompts and responses are discoverable in New York divorce proceedings — including equitable distribution, custody, and maintenance disputes |
Why this matters legally
The February 17, 2026 ruling confirms a principle New York practitioners have warned about for two years: attorney-client privilege under CPLR § 4503 protects communications between a client and a licensed attorney made in confidence for the purpose of obtaining legal advice. A large language model is not a lawyer. OpenAI, Anthropic, and Google are commercial third parties, and disclosing case facts to them destroys any confidentiality interest that could have supported a privilege claim.
The court also rejected the argument that AI outputs qualify as attorney work product under CPLR § 3101(c). Work product protection applies to materials prepared by an attorney or agent in anticipation of litigation. A chatbot is neither retained counsel nor a directed agent — it is a commercial service that retains prompts, trains on user inputs under its terms of service, and produces outputs no lawyer reviewed. Once the bench drew that line, the 31 documents became standard discoverable material under the proportionality analysis in CPLR § 3101(a).
How New York law handles this
New York divorce discovery is broad. Under CPLR § 3101(a), parties must produce "all matter material and necessary in the prosecution or defense of an action," a standard the Court of Appeals has interpreted liberally since Allen v. Crowell-Collier (1968). That reach easily captures ChatGPT transcripts that reveal a spouse's true settlement floor, hidden-asset concerns, or parenting-plan preferences.
Equitable distribution under N.Y. Dom. Rel. Law § 236(B)(5) requires full financial disclosure through the sworn Statement of Net Worth. If a litigant typed into Claude, "I have a crypto wallet worth $180,000 my wife doesn't know about, how do I structure the settlement," that prompt is now producible — and directly contradicts the § 236 disclosure obligation. Courts have sanctioned litigants under CPLR § 3126 for less.
Maintenance calculations under DRL § 236(B)(6) and child support under DRL § 240(1-b) turn on income, which AI prompts routinely expose. Custody determinations under DRL § 240 apply a best-interests standard; a chat log showing a parent asking ChatGPT how to "win 100% custody" or "document the other parent looking unfit" is devastating impeachment evidence that opposing counsel will use at trial.
The spoliation rule compounds the risk. Under Pegasus Aviation v. Varig Logistica (2015), once litigation is reasonably anticipated, a duty to preserve attaches. Deleting ChatGPT history after a divorce filing — or after receiving a hold letter — can trigger an adverse inference instruction under CPLR § 3126. The safer rule is simple: do not create the record in the first place.
Practical takeaways
- Stop inputting case facts into consumer AI tools immediately. This includes ChatGPT, Claude, Gemini, Copilot, Perplexity, and any integrated assistant in Gmail, Notion, or Slack. Assume every prompt is producible in discovery.
- Do not use AI for settlement modeling, hidden-asset questions, custody strategy, or drafting messages to your spouse. These prompts create contemporaneous evidence of intent that opposing counsel can introduce at trial.
- Route all legal questions through your attorney. Communications with New York-licensed counsel made in confidence for legal advice are protected under CPLR § 4503. Your attorney may use AI internally — under enterprise agreements with no-training clauses — but that is a different analysis.
- Preserve existing AI chat history. Do not delete ChatGPT or Claude logs once you anticipate divorce. Spoliation sanctions under CPLR § 3126 can produce adverse inference instructions that are more damaging than the content itself.
- Disclose prior AI use to your attorney. Your lawyer needs to know what prompts exist before opposing counsel finds them through a subpoena to OpenAI or Anthropic, which respond to valid civil process.
- Use enterprise AI tools only with written no-training, no-retention agreements. Consumer tiers of ChatGPT, Claude, and Gemini retain inputs under their terms of service. Enterprise contracts can change that — but only your attorney should be signing them.
Frequently asked questions
Can opposing counsel actually subpoena my ChatGPT history in a New York divorce?
Yes. Under CPLR § 3101(a), broad discovery reaches any material and necessary evidence, and the February 17, 2026 Manhattan ruling confirmed AI chatbot conversations are not privileged. OpenAI retains conversations by default on free and Plus tiers and responds to valid civil subpoenas.
Does using Claude or ChatGPT destroy attorney-client privilege I already have with my lawyer?
No — using AI tools separately does not waive privilege over your attorney communications. However, if you paste your attorney's email into ChatGPT or forward privileged advice to a chatbot, that specific disclosure can waive privilege over those communications under CPLR § 4503 and the common-law waiver doctrine.
What if I already typed case details into ChatGPT before my divorce was filed?
Do not delete the history. Disclose it to your New York divorce attorney immediately so they can assess exposure and address it in your litigation strategy. Deleting prompts after anticipating litigation can trigger spoliation sanctions under CPLR § 3126, including adverse inference instructions at trial.
Are enterprise versions of ChatGPT or Claude safer for divorce-related use?
Enterprise tiers with signed no-training and no-retention agreements materially reduce data-retention risk, but they still do not create attorney-client privilege. Only communications with a licensed New York attorney qualify under CPLR § 4503. Enterprise AI should be reviewed by your lawyer before any case-related use.
Can my spouse get my Claude conversations from Anthropic directly?
Yes, through a properly issued subpoena duces tecum under CPLR § 3120. Anthropic, OpenAI, and Google respond to valid civil process in New York divorce litigation. The February 17, 2026 ruling established that no privilege or work product claim shields those productions absent unusual facts.
Next steps
If you are considering or navigating a New York divorce and have used ChatGPT, Claude, or any AI tool to discuss your finances, custody situation, or settlement goals, consult a New York matrimonial attorney before taking any further action — including deletion. A licensed attorney can evaluate exposure, issue preservation guidance, and communicate with opposing counsel on your behalf under the protection of CPLR § 4503.
This article discusses recent news and provides general legal commentary. It does not constitute legal advice. Every case is unique. Consult a qualified family law attorney for advice specific to your situation.