Every Prompt Creates a Record: How AI Can Jeopardize Confidentiality and Attorney-Client Privilege

Artificial Intelligence (AI) has quickly become part of everyday life. People use it to summarize documents, write emails, organize thoughts, brainstorm ideas, and answer questions. For many, asking an AI tool a question is becoming as natural as running a Google® search. And in many situations, that is perfectly reasonable.

But there is one aspect of AI that people often overlook: Every prompt they give to an AI tool may create a record.

From a legal perspective, that matters. A lot.

When people use AI tools, they often share far more information than they realize. They upload contracts, medical records, proprietary business plans, financial information, employment documents, draft agreements, correspondence, detailed descriptions of disputes and even their present sense impressions. Depending on the tool, its settings, its terms, and the circumstances, those inputs and related outputs may be stored, retained, or accessible later. If litigation or an investigation follows, this material may become the subject of preservation requests, discovery demands, or disputes over confidentiality and privilege.

This is not an argument against AI, but rather a call to proceed with caution.  

The Accidental Evidence Problem

Most people think of AI as a conversation. When someone asks AI a question, it feels informal and temporary, almost like thinking aloud, except the digital room may have a filing cabinet, a retention policy and a surprisingly good memory. Mobile applications on portable devices also make use of AI extremely convenient, ready to serve up quick answers to quick prompts. Legally, however, a prompt may end up functioning more like a written statement than a passing thought.  

When using AI, users can be tempted to upload a detailed timeline of business disputes, copies of emails, and financial records. In so doing, they have now generated a detailed record of facts, allegations, concerns and strategy.

Another often-overlooked risk is that AI prompts can reveal far more than isolated facts. Over time, a collection of prompts may paint a detailed picture of a person’s concerns, motivations, and overall approach to a dispute. For example, in a business dispute, a single prompt may seem harmless. But dozens of prompts asking how to avoid contractual obligations, reduce litigation exposure, pressure an opposing party into settlement, or explain away unfavorable documents could be used to argue that the user had a particular plan or bad-faith intent all along.

What feels like brainstorming in real time could become a costly mistake if those conversations are later characterized unfairly or taken out of context. In litigation, even ordinary brainstorming can look more deliberate once someone adds exhibit stickers and submits it into evidence.

A prompt's discoverability turns on the same factors as any other data: relevance, proportionality, privilege and control. What matters practically is that prompts do not simply disappear, and destroying them when litigation is pending or anticipated can trigger independent preservation obligations. When in doubt, consult your lawyer first before clicking “delete.”

Before You Upload Your Lawyer's Email: Think About Privilege First

If your lawyer sends you a confidential strategy memorandum, a draft lawsuit or an opinion letter, do not upload it to an AI tool and ask the tool to analyze and give you a summary. AI may be a capable analyst or seem like a convenient second opinion, but it is not automatically a member of the privilege club.

Attorney-client privilege protects certain confidential communications made for the purpose of obtaining or providing legal advice. Work-product protection may also apply to materials prepared in anticipation of litigation. Both protections depend heavily on the facts, and disclosure to a third party (including those who control your handy chatbot) can generate an argument that confidentiality was lost or that a protection was waived.

Uploading attorney communications, legal strategies, draft pleadings, or attorney work product to an external AI platform or AI mobile app can therefore turn a protected legal discussion into an expensive dispute over whether the protection still applies. The outcome may depend on the platform’s license terms, privacy and retention settings, security features, the user’s reasonable expectations, the purpose of the disclosure and the governing law.

Practical Pause Before You Prompt

Given all of this, what are the safest default and best practices?

• Do not place privileged or confidential legal material into an AI system unless your lawyer has counseled you on the risks associated with that specific use and tool.
• Assume the prompt may be saved. Write nothing you would be uncomfortable seeing quoted in a deposition, filing, or investigation.
• Remove names, account numbers, medical details, proprietary data and other identifiers when they are unnecessary.
• Use only AI tools approved by your organization, and understand the applicable privacy, retention and training settings.
• If a dispute is pending or anticipated, ask counsel before deleting prompts, chat histories, uploads or outputs. Deleting such materials may constitute spoliation of evidence and could result in court sanctions or other adverse consequences.

The Bottom Line

AI can be useful, efficient, and entirely appropriate. But it should not be treated as a private notebook, a closed-door conversation or a substitute for a secure exchange with counsel. Before submitting a prompt, pause and ask:

• What am I disclosing? 
• Where might it go?
• Who might see it later and
• Am I creating a record that I would be comfortable explaining while under oath?

The safest approach is simple: if you would hesitate to post it publicly or share it with an adversary, think twice before sharing it with an AI platform. At the end of the day, one of the most effective privacy tools isn’t a setting or software feature. It’s taking a moment to think before you press “send.” 

Please contact Hannah D. Kays or any member of the Phelps artificial intelligence team if you have any questions or need advice or guidance.