I am no enemy to generative artificial intelligence in the law. I say this even as I am pretty confident it will eventually gobble up more and more of the profession. I’ve been writing about generative AI and the law since 2023, and the tools have only gotten sharper since. Some frontier models can now summarize a mountain of documents, spot issues, and turn out a decent first draft in seconds. Back in 2023 I started writing that this technology can’t be ignored, and that lawyers have an ethical duty to understand both what it can do and what it can’t.¹ That’s truer today than it was then.
That said, if you are not a lawyer and you are asking an AI chatbot like ChatGPT or Claude about your legal matter, you may be putting yourself, and your attorney-client privilege, in danger. Even if you get the right answer!
It makes sense that clients are turning to chatbots with legal questions. They’re cheap, they’re awake at 2 a.m., they don’t bill by the hour, and they sound very sure of themselves. Heck, increasingly, they give you the right answer – or at least not a widely wrong answer. Hallucinations still happen, the danger didn’t go away, but most of the errors I see today are more subtle than that.
Even so, you shouldn’t be asking AI about your legal questions. You should be asking your lawyer to do it for you.
You’re talking to a company, not a lawyer
A consumer chatbot is a third party. Depending on your plan and settings, the company behind it may keep your conversations and use them. Take Anthropic, the company behind Claude. It says Anthropic will keep prompts on individual plans (Free, Pro, and Max) for five years if they let it train on the data. Users who opt out stay on a 30-day retention period. Deleting your chat logs doesn’t always even mean they’re gone. In the New York Times copyright fight, a federal magistrate judge directed OpenAI in May 2025 to preserve and segregate output log data that would otherwise have been deleted. Later, the district judge affirmed an order requiring OpenAI to produce 20 million de-identified ChatGPT logs to the plaintiffs.
Why does all of this matter? Because the law is very jealous about what you get to keep secret in a lawsuit. And your AI chat logs may not be one of those things. The attorney-client privilege protects confidential communications made to get legal advice. Share those secrets with an outsider, under certain circumstances, and the protection can evaporate.
This isn’t a hypothetical anymore. In United States v. Heppner, Judge Rakoff of the Southern District of New York called the issue a question of first impression nationwide. He ruled that a criminal defendant’s written exchanges with Claude were not protected by either the attorney-client privilege or the work product doctrine. The defendant had argued he was preparing to talk to his lawyers. It didn’t matter. The court explained that what counted was whether he intended to get legal advice from Claude, not whether he later shared Claude’s outputs with counsel. Part of the reasoning came straight from the fine print: because Anthropic’s privacy policy allows data collection, training, and disclosure to government regulators, the court found no reasonable expectation of confidentiality.
To be fair, the law isn’t fully settled. In Warner v. Gilbarco, a Michigan federal court held that a pro se plaintiff’s use of generative AI didn’t waive work product protection, reasoning that AI platforms are tools, not persons, and that disclosure to them is not disclosure to an adversary. But if you’re concerned about someone else reading your chat posts, being told “it depends on the facts” or “it depends on the judge” is not a great answer.
My rule of thumb: treat every chatbot conversation as something you might one day see stamped as an exhibit. If you don’t want to worry about this, don’t ask AI for legal advice.
Lawyers play by different rules, and have better tools
Lawyers aren’t magically immune. We have duties of competence and confidentiality under the Colorado Rules of Professional Conduct, and we have to make reasonable efforts to keep client information from getting out. I wrote in 2023 that lawyers should read an AI vendor’s terms carefully and make sure the vendor is required to maintain confidentiality before using it on client matters.¹ But, when a lawyer is involved, there is a stronger possibility of claiming attorney-client privilege (or work product privilege).
Lawyers can use enterprise and commercial AI platforms, including the same frontier models you’d use at home, under very different terms. Anthropic’s consumer training terms, for example, do not apply to services under its Commercial Terms, such as Claude for Work or API use. And, the legal profession has had a long time to get comfortable with cloud storage, online legal research tools, e-mail, and the like. There’s uncertainty here, too, but generally speaking a lawyer has a better chance of keeping generative AI communications confidential than non-lawyers do.
The legal framework matters just as much. When I use AI on your matter, the work is being done by counsel to give you legal advice. That’s the setting privilege and work product were built for. Even Heppner left that door open: the court specifically identified counsel’s direction as the key factor that could have changed the outcome. Rakoff acknowledged that under the Kovel doctrine, which extends privilege to agents like accountants and translators retained by counsel, AI used at counsel’s direction on a platform with contractual confidentiality might preserve privilege.
The bottom line
If you have a legal question, whether for yourself or your Colorado business, and your first instinct is to ask AI, that’s a good instinct! Just route it through me. Tell me what you want analyzed, summarized, or drafted, and I’ll run it through our firm’s tools. You get the benefit of the technology, a lawyer checking the output, and the protections of the attorney-client relationship.
Ask your lawyer to ask the AI. Same answer, minus the discovery problem.
Frequently Asked Questions
Probably not, and you should assume it isn’t. In United States v. Heppner, a federal court in the Southern District of New York ruled that a defendant’s written exchanges with Claude were protected by neither the attorney-client privilege nor the work product doctrine. Another federal court, in Warner v. Gilbarco, held that a pro se plaintiff’s use of generative AI did not waive work product protection, so the law isn’t fully settled.
It can be. Chat logs can be sought in discovery, and deleting them doesn’t always mean they’re gone. In the New York Times copyright case, a federal magistrate judge directed OpenAI to preserve output log data that would otherwise have been deleted, and the district judge later affirmed an order requiring it to produce 20 million de-identified ChatGPT logs.
Not necessarily. A correct answer doesn’t fix the confidentiality problem. A consumer chatbot is a third party, and depending on your plan and settings, the company behind it may keep and use your conversations. Hallucinations also still happen, although most of the errors I see now are more subtle.
There are no guarantees, but the setup matters. Lawyers can use enterprise and commercial AI platforms under contractual confidentiality terms, and the work is done by counsel to give legal advice. In Heppner, the court identified counsel’s direction as a key factor, and acknowledged that under the Kovel doctrine, AI used at counsel’s direction on a platform with contractual confidentiality might preserve privilege.
Ask your lawyer to run it. Tell your attorney what you want analyzed, summarized, or drafted, and they can run it through the firm’s tools, check the output, and keep the work inside the attorney-client relationship. That is how we handle it at Moriarty Underhill Law.
¹ Colin E. Moriarty, “The Legal Ethics of Generative AI—Part 3,” Colorado Lawyer (Oct. 2023), https://cl.cobar.org/features/the-legal-ethics-of-generative-ai-part-3/
This post is general information, not legal advice, and reading it does not create an attorney-client relationship.