AI chatbots are not your lawyer. Protect attorney‑client confidentiality.

Generative artificial intelligence (AI) has crashed into the legal profession, and it’s no longer a novelty.

A 2025 survey of more than 2,800 legal professionals found that 31% personally used generative AI at work, up from 27% the year before.

Mid‑sized firms are racing ahead: 63% have formally adopted generative AI, even though 81% of their leaders worry about reliability and risk.

As these generative tools become ubiquitous, a stark warning has emerged.

Public AI chatbots are not your lawyer and treating them as one can cost you everything.

The case that opened Pandora’s Box

In United States v. Heppner, federal agents seized a criminal defendant’s devices and found 31 documents generated with a consumer AI platform.

The defendant argued that those AI conversations were confidential because he later shared them with counsel.

Judge Jed Rakoff of the Southern District of New York rejected that claim, finding there was “not remotely any basis for any claim of attorney‑client privilege.”

He gave three reasons:

• AI is not legal counsel. A chatbot cannot create an attorney‑client relationship.
• No reasonable expectation of confidentiality. The AI platform’s terms of service allowed the company to collect prompts and share them with third parties, defeating any privacy expectations.
• No counsel direction. Work‑product protection covers documents prepared by or at a lawyer’s direction. This defendant acted on his own; his prompts reflected his thinking, not his attorney’s strategy.

Judge Rakoff concluded that confidentiality is waived the moment confidential content is pasted into a public AI tool. Once a client uploads an attorney’s memo into ChatGPT, the door cannot be closed.

Not every court has gone that far. In Warner v. Gilbarco, a Michigan magistrate judge held that AI tools are “tools, not persons,” and declined to find a waiver of work‑product protection. But the facts were narrow: The plaintiff was acting as her own lawyer, with no confidential attorney communications shared with the AI.

These cases together illustrate how fragile and fact‑dependent confidentiality has become, and clients cannot afford to assume they are protected.

Why lawyers must warn their clients

The risk is not hypothetical. Copying legal advice into ChatGPT for clarity could make an entire memo discoverable in a future lawsuit.

Pasting a lawyer’s letter into a chatbot to “explain it in simpler terms” may sound efficient, but under Judge Rakoff’s logic, it is akin to reading the letter aloud in a crowded elevator.

The problem persists even with paid subscriptions. Many AI providers reserve the right to collect user inputs and train their models on them.

Until you have a contractual guarantee that the platform will not train on or disclose your data, assume that uploading confidential material into a public AI tool waives confidentiality.

Michael A. Head, Esq., is family law attorney at Fagan Law Group in Atlanta. (Courtesy)

The risk is not hypothetical. Copying legal advice into ChatGPT for clarity could make an entire memo discoverable in a future lawsuit.

Pasting a lawyer’s letter into a chatbot to “explain it in simpler terms” may sound efficient, but under Judge Rakoff’s logic, it is akin to reading the letter aloud in a crowded elevator.

The problem persists even with paid subscriptions. Many AI providers reserve the right to collect user inputs and train their models on them.

Until you have a contractual guarantee that the platform will not train on or disclose your data, assume that uploading confidential material into a public AI tool waives confidentiality.

The stakes extend beyond privilege, too: U.S. courts recorded 487 instances of AI‑generated errors in filings during 2025, with licensed attorneys responsible for nearly 38% of them.

Georgia is addressing this proactively. The State Bar’s Generative AI Toolkit, released in July 2025, makes clear that lawyers may ethically use generative AI only when doing so does not violate professional conduct rules, and that confidentiality, competence and supervision are paramount.

Building a privacy‑conscious AI practice

To harness AI’s efficiencies while safeguarding confidentiality, three principles matter most:

1. Set clear rules. Vague directives to “use caution” are not enough. Firms need explicit policies prohibiting anyone from inputting, pasting or uploading legal advice or confidential materials into any AI tool not expressly approved by the legal department.
2. Use enterprise‑grade tools under counsel’s direction. The Heppner court hinted that the outcome might have differed if a lawyer had directed the AI’s use. Enterprise AI platforms can be configured to prevent model training on your prompts and restrict disclosure. A “lawyer‑in‑the‑loop” workflow keeps AI use supervised and documented.
3. Teach a “pause before you paste” culture. Before pasting anything into a chatbot, ask, “Did this come from an attorney, or does it contain confidential information?” If yes, or even maybe, stop and consult counsel first.

Generative AI offers extraordinary efficiency gains, but it also threatens the confidentiality at the heart of attorney‑client relationships.

As Heppner illustrates, public AI platforms should be treated as third parties. Confidential information shared with them may be exposed to adversaries.

For attorneys and clients alike, embrace AI to streamline routine tasks, but never forget that the confidentiality you protect cannot be delegated to a machine.