AI tools like ChatGPT have become the new Swiss Army knife for business: They can draft a pitch deck, brainstorm a product name, write a sales email, and pressure-test a marketing strategy. The temptation is obvious. If the tool can help you think faster, why not feed it the key details?
Because some “details” aren’t just information. They’re assets. And the way you handle them can determine whether the law will protect them later.
Trade secrets live and die by secrecy
If a company treats confidential information casually, shares it widely, stores it in unsecured places, or discloses it to third parties without nondisclosure agreements, then opponents can argue the company didn’t treat it as a secret. And courts may agree.
Trade secret law is practical. There’s no magic stamp that makes something a trade secret. In general terms, information qualifies as a trade secret only if it (1) has economic value from not being generally known and (2) the owner takes reasonable measures to keep it secret. That “reasonable measures” concept is the pressure point for businesses using consumer AI tools.
Now, consider what happens when someone pastes proprietary information into an AI chatbot.
OpenAI’s own public guidance draws a meaningful distinction: Services for individuals may use content to train models unless the user opts out, while services for business (e.g., ChatGPT Enterprise) do not by default train on your information unless the organization explicitly opts in. (OpenAI)
That matters because if your “confidential” information is being used to improve a model, or is retained and reviewed in ways your business can’t fully control, you’ve arguably done the opposite of taking reasonable steps to preserve secrecy. Even if the odds of the information being reproduced for someone else are low, trade secret protection is about control and process, not your optimism. Whether or not trade secret protection remains after such a disclosure would be determined on a case-by-case basis. It’s best not to build evidence against protection by regularly putting secret information into a non-confidential system.
Patents have a clock, and “disclosure” can start it
Patents flip the script. With trade secrets, you want secrecy forever. Your company’s trade secrets may be the lifeblood of your business, the reason you are competitive in your industry. So long as you keep it secret and it continues to meet the definition of trade secret, the law will protect it. With patents, it’s the opposite. You must disclose your invention to obtain exclusive rights to it. But the timing of that disclosure is everything!
Under U.S. law, certain prior public disclosures, through AI, for example, may stop you from later patenting an invention. But if the inventor files a patent application within one year of making a disclosure himself or herself, the invention may still be patentable. 35 U.S.C. § 102(b)(1). What constitutes a prior disclosure of the invention is broadly defined, including information made “otherwise available to the public.”
So, does inputting an invention into an AI tool count as a “disclosure” that starts the one-year clock?
It depends, and the law is still catching up. If your prompt stays confidential because you’re using a business-tier tool with contractual protections and no training by default, that looks less like a public disclosure. If you’re using an individual consumer tool where content may be used for training (or otherwise handled beyond your control), you are drifting closer to something that could be characterized as making the information available outside your organization – a disclosure.
A quick hypothetical
Imagine a founder and inventor of a new biodegradable packaging film. The founder wants a sharp go-to-market plan, so she asks ChatGPT for help: Target industries, pricing narratives, and positioning against incumbents. To get better output, she pastes in the film’s unique formulation details, manufacturing steps, and performance metrics—information her team has treated as highly confidential.
Six months later, a competitor’s marketing lead runs a similar prompt and gets eerily familiar messaging suggestions. Maybe it’s a coincidence. Maybe it’s convergence. Either way, if the founder later sues for trade secret misappropriation, she may face the uncomfortable question: Were “reasonable measures” taken if the “secret recipe” was pasted into a tool not designed for confidentiality?
And if she later wants to patent the formulation, she may be forced into a debate about whether that AI input was effectively a disclosure that should have started the filing clock.
Practical tips for business owners
- Do not assume that consumer-grade chatbots are like your conference room. If you wouldn’t post it publicly, don’t paste it.
- Use business-grade AI configurations where data is not used for training by default, and confirm what your plan actually provides. When you can, opt out of information sharing and using your data for training.
- Strip prompts of identifying details. If using consumer-level AI products, use placeholders (“Supplier A,” “Process Step 3”) and avoid including exact formulas, customer lists, pricing tables, or source code.
- Create an internal AI policy: who can use which tools, for what tasks, and what categories of information are off-limits. These policies support that you’ve taken reasonable measures to protect secrecy. Follow them.
- If you may patent it, file early. A provisional application is often cheaper than litigating over what counted as a “disclosure.”
AI can absolutely make companies more efficient. But if speed causes you to treat proprietary information like disposable input, you may find out later that the law treated it the same way.
This article does not constitute legal advice.
