AI Tools Aren’t Lawyers — Here’s Why That Matters

Two primary doctrines protect the relationship between a client and attorney:

• attorney-client privilege, and
• work-product privilege.

I’m sure all of you have heard of attorney-client privilege; it allows clients to speak with their lawyers honestly and without fear. Many of you may not have heard of the work-product privilege; this protects attorneys (and those working at their direction) so that the documents they prepare (including analyses, research findings, opinions, and conclusions) in connection with litigation is not discoverable by the opposing party.

An AI tool is not your attorney. Even if you pay to use the tool, unless it’s through an enterprise subscription, the AI tool is typically allowed to retain your data and train future models on it, and the company that owns the tool can disclose your inputs in response to discovery, injunction, or other legal process. Courts have been coming to the same conclusion. For instance, Judge Rakoff of the Southern District of New York recently ruled in U.S. v. Heppner that the 31 documents a criminal defendant generated through “conversations” with a consumer version of Anthropic’s Claude were neither privileged nor protected work product, even though the defendant had received information from his lawyers, fed some of it into Claude to develop defense theories, and then forwarded the outputs to his lawyers. The FBI seized the documents during a search of his home, and properly provided them to the federal prosecutors.

Here are the practical takeaways:

• AI prompts and outputs are generally discoverable. If you type it into a consumer AI tool, assume it can be subpoenaed, just like an email you saved in cloud storage.
• A paid subscription does not create privilege. Standard individual plans do not carry the contractual confidentiality protections of enterprise-tier agreements, and even those do not transform an AI tool into a lawyer.
• Pre-litigation “research” can backfire. Using AI on your own to brainstorm legal exposure, draft responses to subpoenas, or stress-test theories can create a written record of your own analysis that opposing counsel or the government may later obtain.
• Sending AI outputs to your lawyer does not protect them. Results generated or documents created before, and outside of, an attorney-client communication generally cannot be cloaked in privilege after the fact.

At goodcounsel, we help clients think through whether and how to use AI tools in ways that protect — rather than undermine — their legal interests. That includes drafting AI use policies and updating documents to include AI restrictions. Contact us if you’d like to talk through what AI use looks like for your business and where the real risks lie.

Categorised as: Artificial intelligence and the practice of law, The Law