By Stuart Kerr, Technology Correspondent, LiveAIWire

AI in divorce court has already produced its first hard legal precedent, and it went against the person using it. A New York federal court ruled this year, for the first time, that conversations with a public AI chatbot are not protected by attorney-client privilege or work product doctrine, even when the user later shared the AI’s output with their own lawyer. The court’s reasoning applies just as directly to a family law dispute as to the federal investigation the case actually involved: the moment privileged content goes into a chatbot whose terms allow the provider to review, retain or train on it, the privilege protecting that content may already be gone.

That ruling matters far beyond one case because AI in divorce court is no longer a hypothetical convenience. Divorcing spouses are using chatbots to organise financial disclosures, draft custody arguments and analyse a spouse’s discovery responses, often without knowing that doing so can hand the other side a subpoena-ready record of their own legal strategy.

Why AI in Divorce Court Creates a Discovery Problem Nobody Warned Clients About

Family law attorneys writing after the New York ruling laid out exactly why the court reached that conclusion. There was no attorney-client relationship between the user and the AI platform, so ordinary privilege protections never applied. There was no reasonable expectation of confidentiality, because the platform’s terms of service allowed the company to collect, train on, or disclose user inputs. And the conversations were not undertaken for the purpose of obtaining legal advice from counsel, since the user initiated them independently and the AI itself disclaimed giving legal advice.

The practical result for anyone going through a divorce is blunt. Every chat about custody strategy, every pasted bank statement, every AI-drafted settlement proposal is now a potential exhibit. Attorneys are advising clients to treat a “pause before you paste” rule as standard practice, and to assume that anything typed into a public chatbot about their case could end up in their spouse’s hands or before a judge.

The Rules Governing Lawyers Are Getting Sharper, Not Softer

The legal profession’s own guardrails around AI in divorce court are tightening at the same time client behaviour is creating new risk. California’s bar regulators approved proposed rule amendments in March 2026 that would, for the first time, carry disciplinary authority rather than advisory weight. The changes touch six existing ethics rules at once: lawyers would need to independently verify any AI-generated output before relying on it in a case, disclose to clients when AI materially affects how a matter is being handled, treat exposing client data to an AI platform as a potential breach of confidentiality, and verify every AI-assisted citation before filing it with a court.

That last requirement exists because the problem is not theoretical. A public tracking database has documented more than 500 instances since early 2025 in which generative AI produced fabricated legal citations that made their way into real US court filings, with sanctions following in numerous cases. Family law, with its high emotional stakes and frequent use of self-represented litigants, is exactly the kind of practice area where a fabricated citation or a mishandled disclosure is likely to surface.

Judges Are Adopting AI Faster Than the Rules Governing It

The caution applied to lawyers and litigants is not always matched by the courts themselves. Los Angeles County gave half a dozen civil court judges access to an AI tool called Learned Hand in early 2026, which distills lengthy motions and can draft tentative rulings in a judge’s own writing style. The tool is already used by court systems in ten states, and the Michigan Supreme Court adopted it the previous year to review applications for permission to appeal.

Court officials stress that judges must review and edit every draft before adopting it. Not everyone is reassured. Los Angeles County District Attorney Nathan Hochman called the use of AI to generate draft rulings “problematic,” and one judge, speaking anonymously because California rules restrict judges from talking to the press, warned that a tentative AI-drafted ruling becomes a psychological reference point a judge may anchor to even after deciding not to adopt it outright.

Family court dockets carry the same caseload pressure driving civil courts toward these tools, and there is no reason the same pressure will not eventually reach custody and equitable distribution rulings, even though family courts have so far moved more cautiously than general civil courts on adopting AI-assisted drafting. Learned Hand’s own developer describes the tool as a “judicial sous chef” rather than a replacement decision-maker, and the company’s fact-checking layer cross-references every generated sentence against its underlying case-law citations, a safeguard specifically built in response to the kind of fabricated-citation incidents that have already produced sanctions elsewhere in the court system.

Why the Accountability Question Looks Familiar

The tension running through AI in divorce court, an AI-generated recommendation shaping a decision a human is nominally still making freely, is not unique to family law. LiveAIWire’s earlier coverage of facial recognition and the accountability gap in law enforcement found that once an algorithmic output enters a decision-making process, it tends to anchor human judgement even when the human retains formal authority to disagree with it, precisely the dynamic the anonymous Los Angeles judge described. The same structural problem shows up whether the algorithm is scoring a facial match or drafting a custody recommendation: the presence of a confident-looking machine output changes how a human evaluates the case in front of them, whether or not anyone intended that effect.

LiveAIWire’s coverage of the AI credit score’s persistent lending gap found a related pattern in financial decision-making: an algorithmic recommendation with a plausible-sounding rationale is trusted more readily than the evidence supporting it always justifies, a risk that applies as much to an AI-modelled asset division as to an AI-modelled credit decision. Family courts piloting AI-generated financial analysis in equitable distribution cases, currently limited to select California and New York proceedings, face exactly that risk, treating a model’s asset valuation as a neutral baseline when it carries the same unexamined assumptions as any other automated financial score.

LiveAIWire’s broader reporting on how much AI should know about citizens found a similar accountability gap in government decision-making generally: systems adopted through procurement rather than public debate tend to become embedded in operational processes well before anyone outside the institution using them has a chance to evaluate whether the tool actually deserves the trust it is receiving. Family courts adopting AI-assisted drafting or analysis face the same structural risk, adopting tools through vendor contracts and pilot programmes rather than through a public rulemaking process most litigants will never see.

What This Means for Anyone Going Through a Divorce Right Now

For someone navigating a divorce or custody dispute today, the practical guidance from family law attorneys is specific rather than general. Do not paste attorney communications, financial records, or case strategy into any public AI tool, regardless of whether it is free or a paid subscription, because a paid tier does not automatically change a platform’s data retention terms. Assume that anything entered into a chatbot about the case could later be produced in discovery. If an AI tool seems genuinely useful for organising a case, raise it with an attorney first, since courts have suggested that AI use directed by counsel, rather than undertaken independently by a client, stands a better chance of preserving privilege.

For family law practitioners, the direction of travel is equally specific. AI in divorce court is moving from an ethical grey area to a set of enforceable obligations, with California’s proposed rules likely to become the model other states follow rather than an outlier. The firms and courts treating AI verification as a standing procedural step, not a one-time policy memo, are the ones least likely to be the subject of the next sanctions order or the next fabricated citation making headlines.

The asymmetry worth watching closely over the next year is between how carefully lawyers and litigants are being told to treat AI in divorce court, and how quickly courts themselves are adopting AI tools with far less public scrutiny attached. A client who pastes a custody strategy into a public chatbot risks having that strategy handed to their spouse in discovery.

A court that adopts an AI drafting tool through a vendor contract, with no requirement to disclose its use to the parties affected by the ruling it helps produce, faces essentially no equivalent transparency obligation at all. Closing that gap, requiring the same disclosure and verification standards of the institution using AI that clients and their lawyers are now being held to, is the next logical step in a debate that has so far focused almost entirely on individual conduct rather than institutional adoption.

About the Author

Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, emerging technology, and their impact on business, society, and everyday life. LiveAIWire publishes original AI journalism every weekday at liveaiwire.com.