Your ChatGPT prompts could become evidence one day

Many of us have become comfortable asking ChatGPT questions, brainstorming ideas or even drafting documents.

But here is an important Intellectual Property lesson: what you type into an AI tool may one day become relevant evidence.

Recent US cases have started an important conversation

In Assini v. Hayward, a New York court refused to order OpenAI to hand over a litigant's ChatGPT history, finding that the prompts and responses formed part of the person's protected litigation strategy. The subpoena was quashed and the AI conversations remained confidential.

However, another US court took a different approach. In Concord Music Group, Inc. v. Anthropic PBC, an important copyright infringement case, the court ordered the production of AI prompts and outputs that were relevant to the issues in dispute.

Don't let the Assini decision create a false sense of security

The law surrounding AI-assisted work is still developing and different courts have already reached different conclusions.

It is only a matter of time before parties increasingly seek access to AI prompt histories during litigation.

Why does this matter for Intellectual Property?

Imagine you are developing a new invention.

Instead of simply asking ChatGPT technical questions, you upload confidential drawings, explain how your prototype works, describe what your competitors are doing or ask:

"How can I modify my product so that it avoids infringing this patent?"

Those prompts could reveal:

How your invention evolved.
What you knew about a competitor's Intellectual Property.
Whether you were aware of potential infringement.
The commercial thinking behind your innovation.
Evidence that could later be relied upon in patent, copyright or trade secret litigation.

Even if a court ultimately decides those prompts are protected, you may still find yourself spending significant time and money arguing that they should remain confidential.

The practical lesson is simple

Treat prompts to Large Language Models the same way you would treat an email sent to a third party.

If you wouldn't be comfortable disclosing confidential technical information, trade secrets, licensing strategy, or infringement discussions in an email, think twice before including them in your prompts.

AI is an incredible productivity tool - but it should be used thoughtfully, particularly when valuable Intellectual Property is involved.

As AI becomes embedded in our daily work, prompt histories may become just another category of evidence that courts are asked to consider.

The best practice is to never treat a Large Language Model as your confidential notebook.

Note: these are US-based court decisions.

Dr Bernard Dippenaar

Executive | IP & Innovation

bdippenaar@ENSafrica.com