Copying Client Facts into a Public AI Is Not Diligence, It's a Breach of NJ RPC 1.6

It’s a deceptively simple act, one that feels like a modern efficiency hack. You copy a thorny paragraph from a client’s email, a confusing clause from an opposing party’s motion, or a set of facts for a demand letter, and paste it into a public AI chatbot with the prompt, “Simplify this,” or “Draft a response.”

For the solo and small firm attorney in New Jersey, this shortcut is a ticking time bomb. It’s not a potential ethics violation or a gray area to be debated. It is a direct and unambiguous collision with your most fundamental duty: confidentiality, as enshrined in New Jersey’s Rule of Professional Conduct 1.6.

The Myth of the 'Anonymous' Prompt

The most common self-justification I hear from attorneys is that they are careful to remove names and obvious identifiers. This provides a false sense of security. The bedrock misunderstanding here is how public-facing large language models (LLMs) operate. When you use a free or consumer-grade AI tool, you are not merely having a private conversation; you are often providing raw material for the AI’s future training.

Read the terms of service for most public AI platforms. They typically reserve the right to use your inputs to refine and improve their services. Your “anonymized” fact pattern—the unique sequence of events, the specific dollar amount in a dispute, the niche corporate structure—becomes a permanent part of the model’s data corpus. This data can be synthesized, recalled, and potentially reconstructed in response to queries from other users down the line.

Think of it this way: discussing a client’s matter in a hushed tone on a packed NJ Transit train from Trenton is a lapse in judgment. But pasting those same facts into a global AI model is like handing a copy of your case file to the train conductor to use as he sees fit. The factual pattern itself is confidential information, and de-identifying it is far more difficult than simply deleting a client’s name.

NJ RPC 1.6: An Unforgiving Standard

New Jersey’s RPC 1.6(a) is expansive: “A lawyer shall not reveal information relating to representation of a client unless the client gives informed consent, the disclosure is impliedly authorized in order to carry out the representation, or the disclosure is permitted by paragraph (b), (c), or (d).”

The key phrase is “information relating to representation.” This is an intentionally broad standard. It covers far more than privileged communications. It includes any and all information gleaned during the course of the attorney-client relationship. That fact pattern you’re asking an AI to analyze is, by definition, “information relating to representation.”

Submitting it to a third-party service whose business model involves absorbing and reusing that information constitutes a “revelation.” And it’s a revelation for which you do not have informed client consent. The exception for being “impliedly authorized” doesn’t apply, as there are now numerous secure methods to achieve the same result without exposing client data to a public training set. Exposing client data for the sake of convenience is not a necessary component of carrying out the representation.

The Ethically Compliant Path Forward

Pointing out the danger is useless without providing a solution. Protecting client confidentiality doesn’t mean abandoning the powerful advantages of AI. It means adopting a professional, secure approach.

1. Mandate Enterprise-Grade, Zero-Retention Tools: The only acceptable AI tools for handling client data are those with explicit, contractually guaranteed “zero-retention” policies. This means the vendor agrees not to store your prompts or completions and, crucially, will never use your firm’s data to train its models. Services like ChatGPT Enterprise or Microsoft Copilot with Commercial Data Protection are built for this. Scrutinize vendor contracts for these specific clauses. They are non-negotiable.

2. Master the Art of the Abstracted Prompt: For quick legal research or brainstorming on public tools, learn to work with hypotheticals. Instead of pasting client specifics, abstract the legal question.

   • Don't: “My client, a bakery in Cherry Hill, was sold expired flour by a Trenton distributor. The contract is silent on remedies for defective goods. What are my client’s options under the NJ UCC?”
   • Do: “What are the remedies for a buyer of commercial goods under the New Jersey Uniform Commercial Code when the seller delivers non-conforming goods and the contract is silent on remedies?” This approach allows you to leverage the AI’s legal knowledge base without revealing a single confidential fact.

Your duty under RPC 1.6 is not merely to avoid direct harm; it is to prevent the unauthorized revelation of information. In the age of AI, that requires a new level of diligence. The convenience of a quick copy-paste is not worth the catastrophic risk of a client data breach and an ethics complaint. Review your workflows today and ensure your firm’s use of AI builds client trust, rather than silently eroding it.