Artificial intelligence can be an incredibly useful tool. It can summarize information, organize thoughts, improve writing, analyze technical concepts, and even assist with preparing patent-related materials. But if you have an invention or valuable confidential business information, there is an important question to ask before typing it into an AI prompt: WHAT RISKS AM I TAKING WITHOUT EVEN REALIZING IT? Using AI with confidential information can create several different risks which can affect patent rights, trade-secret protection, and even who legally qualifies as an inventor.
1. Can Using AI Jeopardize Your Patent Rights?
Patent rights depend heavily on timing and disclosure. Under US patent law, an invention must not have been patented, described in a printed publication, publicly used, offered for sale, or otherwise made available to the public before its effective filing date, subject to certain exceptions. The US provides a limited one-year grace period for certain disclosures originating from the inventor. But relying on that grace period comes with significant risks. Many foreign countries impose much stricter requirements, meaning that a public disclosure before filing may immediately eliminate patent rights in those jurisdictions. So, what happens when an inventor gives the details of an unfiled invention to ChatGPT, Claude, Gemini, or another generative AI platform?
The law is still developing. There does not yet appear to be a reported US patent decision holding that merely entering an invention into an AI prompt necessarily constitutes a “public disclosure” under 35 U.S.C. § 102. But that does not mean it is safe. Depending on the platform, account type, contractual terms, and privacy settings, information submitted to an AI system may be retained, reviewed, used for training or system improvement, or otherwise made accessible outside the inventor's control.
The risks are simply too significant to assume that an AI system will keep your information confidential. DO NOT ENTER CONFIDENTIAL OR PROPRIETARY INVENTION INFORMATION INTO AN AI PLATFORM UNLESS YOU ARE CERTAIN THAT THE SYSTEM IS SUBJECT TO APPROPRIATE CONFIDENTIALITY, DATA-SECURITY, AND NON-TRAINING PROTECTIONS THAT PRESERVE THE CONFIDENTIALITY OF YOUR INFORMATION.
2. Trade Secrets May Be Even More Vulnerable
Trade-secret protection depends on secrecy. Under the federal Defend Trade Secrets Act, protected information must derive economic value from not being generally known, and its owner must take reasonable measures to keep that information secret. That can become difficult if the information is voluntarily provided to an outside AI provider that has no obligation to keep it confidential. A recent 2026 federal case illustrates the concern. In Trinidad v. OpenAI, Inc., a plaintiff asserted trade-secret rights in proprietary frameworks she had developed using ChatGPT. The court dismissed her federal trade secret claim, reasoning that she had voluntarily shared the alleged trade secret information with OpenAI and accepted its terms of service without negotiating an obligation requiring OpenAI to maintain its confidentiality. Again, companies should be cautious before placing proprietary formulas, source code, product specifications, customer information, pricing models, strategic plans, or other confidential information into third-party AI systems.
3. Courts Are Already Questioning Whether AI Conversations Are Confidential
A separate 2026 federal decision provides another cautionary example. In United States v. Heppner, the defendant used Anthropic's Claude to help develop materials relating to his criminal defense and later claimed that the resulting communications were protected by attorney-client privilege or the work product doctrine. The court disagreed. The court examined the AI provider's privacy policies and concluded that the defendant did not have a reasonable expectation of confidentiality in his communications with the AI platform. Heppner is not a patent case, and it does not hold that giving an invention to an AI platform constitutes a public disclosure under patent law. But its reasoning reinforces an important principle: DO NOT AUTOMATICALLY ASSUME THAT INFORMATION ENTERED INTO AN AI PROMPT IS CONFIDENTIAL.
4. There Is Another Patent Risk: Who Actually Invented It?
AI presents another issue that is distinct from confidentiality, namely inventorship. Under US patent law, an inventor must be a natural person. AI itself cannot be named as an inventor. The Federal Circuit confirmed that principle in the 2022 Thaler v. Vidal decision. The more difficult question arises when a human inventor uses generative AI during the inventive process. Using AI does not automatically make an invention unpatentable. A human inventor can use AI as a tool. The important question is who conceived the claimed invention? If the human inventor conceived the inventive concept and uses AI merely to assist, the use of AI may not create an inventorship problem. But the situation becomes more complicated if the inventor gives an AI system a problem and the AI independently proposes the inventive solution. A human must contribute to conception, and an invention conceived solely by AI cannot simply be attributed to a human user. For that reason, inventors using AI during product development should maintain records including notebooks and AI logs, showing what the human inventor conceived, what was provided to the AI, and what the AI contributed.
5. Not All AI Systems Are the Same
It is also important not to lump every AI system together. Consumer-facing AI products, paid business accounts, enterprise platforms, privately hosted models, and closed internal systems may provide very different protections. Before putting confidential IP into any AI system, consider:
- whether prompts and uploaded documents are retained;
- whether information can be used for model training or improvement;
- whether humans can review the information;
- whether the provider can disclose information to third parties;
- whether contractual confidentiality obligations exist;
- who owns or controls the inputs and outputs; and
- whether the company has the ability to prevent its data from being incorporated into a public model.
The distinction between public and secured AI models can be critical. Simply turning off a “use my data to improve the model” setting may reduce risk, but it should not automatically be assumed to resolve every confidentiality, trade-secret, or patent issue.
6. A Practical Rule for Inventors: Protect First, Prompt Second
AI can be a powerful tool for inventors and businesses. The point is not that AI should never be used. The point is that valuable intellectual property should not be casually placed into an AI system without understanding the consequences. Before asking AI to draft a patent application, analyze an unfiled invention, solve a technical problem, improve confidential technical language, or develop a proprietary product concept: PROTECT THE IP FIRST. That may mean filing a patent application before disclosing the invention, using an AI environment with appropriate contractual and technical confidentiality protections, limiting what information is provided to the system, and documenting the human inventor's contribution to conception. Saving time with AI can be valuable. Accidentally jeopardizing patent rights, trade-secret protection, or inventorship is not.