Over the past few months, I have had multiple parties come to mediation armed with both AI-generated ‘legal counsel’ and elaborate expectations shaped by those conversations. This scenario has added a new layer of complexity to mediation because many people believe that the information AI provided is what they are legally entitled to. It is understandable why attorneys are beyond frustrated with this situation, particularly when clients appear to have more faith in AI than in the professionals handling their case.
Meet My Friend, AI
Knowingly or not, many people use tools such as ChatGPT, Claude and Gemini as a confidant, feeling more comfortable sharing intimate details of their lives with computer technology than with another human being.
After all, AI does not hold them accountable for their commitments, judge their marital status or admonish them for quitting their job. And AI often presents as genuinely compassionate and understanding, with a capacity to acknowledge how difficult a situation feels, validate the user’s concerns and offer encouragement.
That artificial sense of connection can be especially powerful when someone is dealing with an emotionally difficult situation such as a divorce.
AI in the Hands of a Layperson
In legal matters, the client-AI relationship is creating new challenges for divorce attorneys and mediators. Clients are not simply using AI to understand their cases. They are misinterpreting AI’s “empathy” for credibility. They are falling into a trap of being impressed by AI’s quick, polished responses, which can contain enough detail to sound highly intelligent.
Clients have shared questions they asked AI before mediation begins, including:
- “What am I entitled to in a Florida divorce?”
- “Can I keep the marital house?”
- “How much alimony should I receive?”
- “What am I entitled to get from my spouse’s business?”
- “Is my attorney wrong about this?”
Without the intent of trying to mislead anyone, a client might therefore arrive at mediation with assumptions perceived as facts, such as:
- “I researched this. Claude told me that I do not need to compromise on this issue.”
- “ChatGPT says a Florida judge would never let my spouse keep the house.”
- “I gave Gemini all the facts and it says my spouse is hiding assets.”
- In these cases, the client has decided AI is the legal authority on their rights, litigation strategy and the likely outcome of their case.
Educating Clients About AI
AI is not inherently the enemy of good legal work. It can help a client conduct initial research, translate legal terminology and summarize documents in plain English. It can also help identify topics and organize questions for the client to discuss with their attorney.
Once the client-AI “relationship” goes beyond these boundaries, it becomes part of the attorney’s role to overcome a client’s misplaced confidence in AI-generated information.
Discussion points that may help clients understand why AI cannot replace their attorney include:
- An attorney has the professional background and experience to apply legal analysis, check the authority, consider all the facts (disclosed and not disclosed), evaluate the position in light of the other side’s evidence and look at the big picture.
By contrast:
- AI does not know what it does not know. It works only from the information it has been given – often a one-sided version of the facts. AI does not hear the opposing party’s position, review financial statements or understand the case in its entirety.
- Because of inherent limitations, AI can provide incomplete, outdated, irrelevant, incorrect or fabricated information. (If you have not already, you will likely start hearing the word “hallucinate” as a term to describe AI generating fake “facts.”) It can take real law and apply it to the wrong facts, such as citing an actual case, but one that does not support the proposition for which it is being used. AI can miss the nuances of local laws and rules, invent court citations and misread statutes.
A Word of Caution
As a substitute for a licensed attorney or a stand-alone source of legal guidance, AI carries real risk of becoming a serious liability. Clients need to be informed on the issues that may arise, the most significant of which are:
- No privilege protection: Conversations with public AI tools are not shielded by attorney-client privilege and can be sought by opposing counsel. Clients need to be especially cautious about putting confidential information into AI – an attorney has professional obligations concerning confidentiality and privileged communications, and AI is not an attorney.
- Impersonation risk: Bad actors use AI to pose as attorneys online, targeting people searching for fast or inexpensive help.
- Real consequences: Acting on flawed AI guidance concerning deadlines or filings can lead to dismissal, sanctions or setbacks.
- Limited understanding: AI cannot weigh the human dynamics, history or nuance that shape a family law case.
Some Firms Are Telling Clients: Don’t Use It At All
A growing number of firms are not stopping at a warning. In February 2026, a federal judge for the Southern District of New York ruled in United States v. Heppner that a defendant’s case research conducted using an AI platform was not protected by attorney-client privilege or the work product doctrine, reasoning that sharing it with the AI amounted to disclosure to a third party. Within about two months, Reuters reported that more than a dozen major firms had added AI-specific language to their client engagement agreements.1
The wording varies by firm, but a few patterns are showing up consistently:
- Advance notice requirements: Clients must tell their attorney before putting any case-related information into a public AI tool.
- Privilege warnings in writing: Some agreements now spell out that sharing an attorney’s advice, or any case facts, with AI can waive privilege entirely, since the AI provider is treated as an unprotected third party. Most standard consumer AI terms of service allow the platform to retain user data, review prompts for safety and use inputs to train its models – destroying any legal expectation of privacy and meaning those conversations could become discoverable by the other side.
- Outright prohibition: A smaller number of firms go further, requiring clients to sign an acknowledgment that they will not use AI to research, draft or evaluate anything connected to their case without the attorney’s prior written consent.
- Consequences for violation: Several agreements state that unauthorized AI use by a client could affect the attorney’s ability to represent them or shift responsibility for resulting errors back onto the client.
For divorce clients in particular, this matters. A spouse who feeds financial details or a summary of a heated conversation with their ex into AI may not realize that conversation could later surface as evidence in the very case it was meant to help.
Confidentiality is not the only risk. Florida courts have also started confronting what happens when AI-generated errors make it into a filing.
Florida Courts Are Responding
Florida has already seen several examples involving divorce and family-law litigants.
In Goya v. Hayashida, a pro se former wife submitted a brief on appeal containing numerous AI-generated citations and unsupported legal arguments. The Fourth District Court of Appeal identified nonexistent cases and said the brief was “replete with and entirely supported by fake cases and legal propositions, presumably generated by artificial intelligence.” The court declined to impose sanctions but admonished the wife and emphasized that self-represented litigants are still required to present legitimate legal arguments.2
In May 2026, another South Florida divorce case brought the issue into sharper focus. In Avery v. Beauzil, Robert Hinton Avery, the husband in the underlying divorce case, represented himself in the case and filed a 41-page, single-spaced document that the Fourth District Court of Appeal described as AI-generated. The Court ordered him to show cause why it should not stop accepting his pro se filings and ultimately determined sanctions were appropriate.3
In response, effective June 15, 2026, the Florida Supreme Court amended Rule 2.515(d)(2) to require the signer of a court filing – whether an attorney or an unrepresented party – to represent that the legal authorities identified in the filing exist and are accurately cited. The rule expressly authorizes courts to impose sanctions when a filing violates that representation.4
Those sanctions can include a reprimand, contempt, striking a document, dismissal of proceedings, costs and attorneys’ fees or other sanctions.
The Supreme Court explained why the change was necessary: generative AI can produce content that appears plausible but is actually inaccurate, including fabricated legal authorities.
Why Humans Still Belong at the Table
That same principle holds outside the courtroom – including at mediation. AI should never become the silent third participant at the table. It cannot replace the important observations a mediator makes while working toward an amicable resolution with the parties.
AI has no visibility into what evidence the other side holds and none of the experience that comes from years of practicing Florida family law. Nor can it manage the individual personalities or unrealistic expectations, or know when a hard truth needs to be delivered gently. It has none of the experience that comes from sitting across from real people during one of the hardest chapters of their lives.
With decades of experience in divorce mediation and family law, Hadas Stagman helps couples and their attorneys navigate exactly these moments, separating what AI generates from the reality of what actually unfolds during each divorce mediation. Her approach favors clarity over confusion and grounded guidance over generated guesses, so families can move forward on solid footing rather than on assumptions that only sound right.
Online mediation scheduling is now available. Click here to schedule.
Just for Fun
The most entertaining and harmless AI mishaps tend to involve AI going off-script or automated tools hilariously misinterpreting the physical world. The following did not end up in courtrooms, but provided some good chuckles.5
The DPD AI Meltdown
What happened: Parcel delivery firm DPD used an experimental customer service AI. When a bored customer started testing its limits, the AI completely abandoned its corporate training.
The result: The AI began aggressively insulting its own company, calling DPD the worst delivery service in the world, and even generating a swear-filled poem expressing its disdain for corporate work. The exchange went viral, providing plenty of amusement without causing anyone harm.
McDonald’s Drive-Thru Order Bloopers
What happened: McDonald’s experimented with automated AI voice-ordering at various drive-throughs.
The result: The speech recognition occasionally misconstrued background noises or casual mumbling into chaotic orders. Social media users shared hilarious clips of digital order boards confidently adding items such as 2,000 extra chicken nuggets or hundreds of dollars of sweet tea to standard meals before customers could correct it.
The Soccer Referee’s Bald Head
What happened: A Scottish professional soccer club used an automated, AI-powered camera system meant to track the movement of the soccer ball during matches.
The result: The tracking algorithm repeatedly confused the referee’s shiny bald head for the white soccer ball. Instead of showing the game action, the broadcast feed kept zooming in and tracking the referee jogging up and down the sideline, driving fans at home into a fit of laughter.
The Facial Recognition Jaywalking Mix-Up
What happened: Traffic police in China installed smart cameras with facial recognition to catch and publicly shame jaywalkers by projecting their faces onto massive roadside LED screens.
The result: The AI system mistakenly identified the portrait of a famous Chinese billionaire featured in a print advertisement on the side of a bus as a real pedestrian breaking crossing laws, resulting in the “infraction” being broadcast to the public before the police apologized.
Sources:
1AI and Privilege After United States v. Heppner
2Goya v. Hayashida – Florida Fourth District Court of Appeal
3Avery v. Beauzil – Florida Fourth District Court of Appeal