When Your Client Fact-Checks You With AI: Confidentiality, Privilege, and Professional Judgment

A few years ago, almost none of my clients used AI when they communicated with me. Today, nearly all of them do.

I research, analyze, and plan for a client’s business, then send my advice. Often the next thing that happens is predictable. The client pastes my work into a public AI tool and asks it to summarize, critique, or improve on what I wrote. Then I get a reply that begins, “I ran this through AI, and it says…”

If you are an attorney, accountant, consultant, financial advisor, or any other professional, you have probably received that email. This article covers two things. The first is why that habit is riskier for the client than most people realize. The second is how to handle it well, because it isn’t going away.

The Problem Is Bigger Than Annoyance

I’ll admit this is frustrating. I spend hours on legal research and strategy, and I draw on more than 20 years of experience. Then I’m asked to defend that work against a chatbot that has never seen the contract, the course of dealing, or the people involved.

Frustration isn’t the interesting part. Three real problems sit underneath it.

1. Confidentiality may be compromised. Clients paste privileged memoranda, strategy, and sensitive financial or business information into tools they don’t control.
2. The AI is often wrong in ways that are hard to see. General-purpose models can miss jurisdiction-specific rules and the facts that actually drive the outcome. They also tend to sound confident while being wrong.
3. The relationship can erode. When clients act as their own second-chair professional, trust and efficiency both suffer.

Why Feeding Advice Into AI Can Undermine Confidentiality and Privilege

This is the point I most want clients to understand, so I’ll state it carefully.

The attorney-client privilege protects confidential communications made for the purpose of getting legal advice. In Florida, the privilege is codified in section 90.502 of the Evidence Code. The word confidential matters a great deal. Under section 90.507, a privilege can be waived when the holder voluntarily discloses a significant part of the privileged matter. Accountants have a similar statutory privilege in Florida under section 90.5055, with similar vulnerabilities.

The privilege belongs to the client, not to the lawyer. So when a client pastes my confidential memorandum into a public AI platform, the client may be making a voluntary disclosure to a third party. Many consumer AI tools:

• Store prompts and outputs on company servers.
• May allow human review of conversations for safety or quality purposes.
• May use inputs for model improvement unless the user opts out.
• Have privacy and confidentiality terms that vary by product, plan and settings.

Courts are only beginning to address this. In early 2026, a federal court in New York (United States v. Heppner) reportedly held that a defendant’s exchanges with a public AI platform were not protected by attorney-client privilege or the work product doctrine. The direction of travel is clear enough. Treat an AI platform as a stranger to the conversation unless you have strong reason to think otherwise.

Two caveats keep me honest:

• The law is still developing. Whether a given AI interaction waives privilege depends on the platform, its terms, the client’s settings, and the facts. Enterprise tools with contractual confidentiality protections are a different category from a free consumer chatbot.
• Work product has its own rules. Waiver of work-product protection generally follows a different standard than waiver of privilege, so the analysis isn’t identical for every document.

My practical advice doesn’t change either way. Anything a client puts into an unvetted public AI tool should be treated as potentially no longer confidential.

The same logic applies well beyond law. It covers an accountant’s tax strategy, a consultant’s proprietary analysis, a financial planner’s recommendations, and an HR advisor’s investigation notes. Professionals in every field have a stake in this.

Why the AI Answer Is Often Wrong

I don’t dismiss what clients bring me. Sometimes the AI raises a fair question. More often it goes off track in one of these ways:

• Missing context. The model doesn’t know the facts that aren’t in the prompt. In litigation and in deals, those missing facts are often decisive.
• Wrong jurisdiction or outdated law. A general model may blend rules from different states or rely on law that has changed.
• Fabricated or unreliable citations. AI-generated “authority” must be verified, and it sometimes does not exist.
• Leading prompts. If the client frames the question around the answer they hope for, the AI will often agree.
• No accountability. The AI carries no license, no malpractice exposure, and no duty to the client. I do.

How to Handle It: Practical Pointers for Professionals

1. Don’t dismiss it, and don’t take it personally

Clients who use AI are usually trying to be informed participants. Treating it as an insult costs you trust. I read what they send, and I think that matters. Sometimes the AI catches a point I want to examine again. Sometimes it’s wrong, and I can explain why. In both cases, taking it seriously earns credibility.

2. Address confidentiality at the start, in writing

Put it in your engagement letter and onboarding materials, in plain language. Tell clients that:

• Your advice and work product are confidential and intended for them alone.
• Putting them into public AI tools may jeopardize privilege and confidentiality.
• They should check with you before using AI on anything sensitive.

Telling a client once, verbally, is not enough. People forget, and a written policy supports you later.

3. Explain the “why,” not just the “don’t”

Clients who hear “never use AI” tend to ignore you. Clients who understand the consequence behave differently. I tell them: “If this later becomes part of a dispute, the other side may argue you waived the protection that kept our communications private. I’d rather not give them that argument.”

4. Give them a better path

Because AI use is a given, offer a controlled alternative:

• Invite them to send you their AI-generated questions directly.
• Offer a short call to review whatever the AI suggested.
• If appropriate, recommend vetted tools with contractual confidentiality protections, and have them anonymize facts before using anything else.

5. Respond to AI-generated pushback with substance

When a client brings an AI critique, I do the following:

• Acknowledge the question.
• Separate what is correct from what is not.
• Explain the reasoning, the facts, the jurisdiction-specific rules, and the strategy the AI did not have.
• Document the exchange, particularly when the client decides to go against my advice.

Documenting is how you protect yourself when the AI-driven suggestion turns out badly.

6. Keep exercising independent professional judgment

AI can be an input. It is never a substitute for the professional’s duty to the client. I verify the research, I analyze the facts, and I own the advice. Your own professional obligations, such as those in Florida Rule of Professional Conduct 4-1.6 on confidentiality and the Florida Bar’s guidance on generative AI, apply to your own use of AI tools as well. Hold yourself to the standard you ask of clients.

7. Set boundaries on scope and fees

Reviewing AI output takes time. Set expectations early about whether that review is included or billed separately, and communicate this transparently. Being open about it prevents resentment on both sides.

8. Treat it as part of the process

I’ve settled on this view: AI-assisted clients are the new normal. The professionals who thrive will be the ones who build AI into their client communication, rather than fighting it. Educate early, listen respectfully, correct with substance, and protect confidentiality.

A Checklist You Can Use Today

• Add an AI and confidentiality clause to your engagement letter.
• Send existing clients a short written advisory about AI and privilege.
• Mark privileged communications clearly.
• Offer a standard process for reviewing AI-generated questions.
• Document client decisions that go against your advice.
• Review your own firm’s AI tools and policies for confidentiality compliance.

Frequently Asked Questions

Does using ChatGPT or another AI tool waive attorney-client privilege?

It can. Privilege depends on confidentiality, and voluntarily sharing privileged communications with a third-party platform risks waiver. The outcome depends on the tool, its terms, and the facts, and courts are still working through the issue.

Is it safe to paste my lawyer’s advice into AI?

Generally not on a public consumer tool. Ask your attorney first, and never paste in anything you wouldn’t want an adversary to read.

What should a professional do when a client says “AI says you’re wrong”?

Listen, evaluate the point honestly, explain what the AI missed or got right, and document the conversation.

Can AI replace my attorney or accountant?

No. AI can help you ask better questions. It doesn’t know your full facts, doesn’t carry professional accountability, and isn’t responsible for the outcome.

Final Thoughts

The goal isn’t to push AI away. The goal is to make sure it strengthens the professional relationship instead of weakening it. When clients understand how privilege works, and professionals respond calmly and substantively, AI becomes one more tool in a collaborative, strategic process.

If you are a business owner with questions about protecting confidential communications, or a professional wanting to build an AI policy for your clients, contact Fornaro Legal at 954-324-3651 or schedule a consultation. We serve businesses across Coral Springs, Parkland, Broward County, and South Florida.

This article is for general informational purposes and is not legal advice. Reading it does not create an attorney-client relationship.