You May Want to Keep Some Secrets From the Robots

Let’s talk about Large Language Models (LLMs)—you know, those remarkably helpful AI tools that can draft your emails, refine your pitch deck, (maybe even help you write a blog post…) and, in certain cases, have the unintended consequence of undermining your patent strategy.  A true multitasker.

If you’re working on a new invention or idea, you’ve probably considered pasting your idea into an AI tool to “see what it thinks.” Maybe you wanted help sharpening the concept, drafting a description, or stress-testing your thinking. If nothing else, your friendly AI bot is bound to tell you what a great idea you’ve got so that you’ll like it even more!

The Core Issue (Without the Legal Lecture)

Patent systems around the world share a charming quirk: they tend to value novelty. More specifically, they often require that your invention not be disclosed to the public before you file for protection or, in the US, not more than one year before such a filing. “Public,” incidentally, does not require a press release or a keynote address. It can include quieter forms of disclosure—like sharing details with third parties without clear confidentiality protections. Even if it isn’t patentable, a trade secret only retains its protection under applicable law by being retained as a secret…

While it may not be intuitive, when you input proprietary information into an LLM, you may:

• Considered to have shared it publicly;
• Have agreed to terms you didn’t read (no judgment; you’re in good company);
• Potentially permitted that information to be stored, reviewed, or otherwise used.

These outcomes may not be aligned with a pristine IP strategy.

A Real-World Reminder (Subtle, but Memorable)

In 2023, Samsung reportedly discovered that its engineers had been uploading confidential source code and internal materials into a public AI chatbot to troubleshoot problems and help debug the code. Efficient? Arguably. Ideal from an information-control standpoint? Not especially. Samsung responded by restricting the use of such tools, which is typically not a move companies make when everything is going just the way they wanted.

While that incident wasn’t specifically about patents, the principle holds: valuable information has a way of escaping its intended boundaries when shared with systems you don’t fully control. Replace “source code” with “pre-filing invention details,” and the outcome becomes more legally interesting.

But Surely It’s Private, Right?

Sometimes. Sometimes not. Some AI platforms offer enterprise-grade confidentiality commitments. Others make more limited promises. And some rely on terms of service that are, let’s say, character-building to read. If your working assumption is “this is probably fine,” it may be worth upgrading that to “this is explicitly protected.”

Because if a disclosure isn’t confidential, it may be treated—legally speaking—as… not confidential. While putting your proprietary information into a LLM does not guarantee disaster, it simply introduces risk that could have been avoided with modest planning—something about which entrepreneurs are not always known to be enthusiastic.

Practical Guidance (Efficient and Slightly Unexciting)

• File first, then experiment. A provisional patent application is often faster and less burdensome than explaining to investors why your IP position became… nuanced.
• Use the right tools. If AI is part of your workflow (and it probably should be), consider versions with clear, contractual confidentiality protections.
• Abstract when possible. Asking for help at a conceptual level is generally safer than providing a fully operational blueprint.
• Skim the terms. Or assign it to someone who enjoys that sort of thing and hasn’t complained recently.

The Bottom Line

AI tools can be great at helping to accelerate work, improve clarity, and make you look impressively efficient. They are not, however, your attorney, your confidentiality framework, or your risk manager... I’m not telling you not to use AI, just saying that, with respect to your IP, you might want to use it with enough restraint to ensure your next funding round isn’t accompanied by an unexpectedly philosophical discussion about “what constitutes public disclosure” and whether that bot was appropriately concerned about your patent portfolio.