Imagine That IP Law Blog

Increasingly, public AI tools that appear to be impressive are tempting to use for research, technology development, document preparation, and general queries on a variety of consequential issues including law in general, and patents in particular.  However, submitting information to a public AI tool can result in several overlapping dangers, such as patentability problems, loss of trade secret protections, breach of confidentiality obligations, and even export-control violations. 

Below, this article outlines some of the dangers that individuals, companies, and law firm clients should consider before inputting any confidential information regarding an invention into any external, public AI platform.

United States patent rights may be destroyed
The United States patent system rewards novelty, and a patent application’s filing date is the key date against which novelty is measured.  If an invention is no longer new, for example, by having already been made available to the public before filing a patent application, such disclosure may become “prior art” that can be used to reject or invalidate the claims of a later-filed patent application.

One danger with using a public AI tool is the uncertainty regarding whether the prompts are a “public disclosure” in some way.  Some critical questions arise. Does the prompt remain private? Is it retained by the third-party system? Can it be reviewed by humans? Will it be used to train the AI tool itself? Could the subject matter manifest in another user’s output? Could the information be shared with a third-party plugin, subcontractor, or analytics provider?

A prompt entered into an AI tool is not private to the user unless the AI tool’s governing terms, architecture, and controls are effective in making it so. If the information is accessible to people or systems that are not bound by confidentiality provisions or part of a closed or “sandboxed” environment, a party seeking to invalidate a patent may argue that the invention was made available to the public.

The United States provides some exceptions for inventor-originated disclosures made one year or less before the effective filing date, but that grace period is not a guarantee.  At best, the grace period is a fallback position – proving the facts necessary to avail of the grace period may be burdensome, expensive, and not guaranteed to succeed.

Foreign patent rights may be destroyed
Even if the United States and a few other jurisdictions around the globe provide for a one-year grace period for public disclosures before filing a patent application under some circumstances, some foreign patent systems do not. These are so-called “absolute novelty” jurisdictions. Europe is a typical example, as is China.  In those jurisdictions, aside from narrow exceptions such as evident abuse or certain official exhibitions, a pre-filing disclosure can be cited against the applicant and bar patent protection even if the applicant themselves made the disclosure.

This means that while an inventor pasting a detailed invention description into a public AI platform before filing may still be able to attempt a U.S. patent filing (under the exception described above), they may have fatally compromised their European rights (and possibly rights in other jurisdictions). This may be commercially devastating, since patent value may depend on international coverage. Investors, licensees, and patent purchasers are often not solely interested in the U.S. and may ask whether you have also preserved the option to file patent applications in other major markets.

Trade-secret protection can be lost
Trade secrets may be protectable if they derive value from not being generally known and if reasonable steps are taken to keep them secret. Feeding technical details into a public AI tool can undermine this protection. This matters especially because not every idea or invention can obtain patent protection. For example, some manufacturing methods, data-processing pipelines, formulas, know-how, tuning parameters, test data, source code, and process optimizations may be more valuable when protected as trade secrets. Once “secret” details are submitted to an uncontrolled public AI tool, a company may struggle to show that it took the reasonable secrecy measures needed to claim trade secret protection. Even if the AI provider never leaks the information, the act of disclosure to a third party without adequate confidentiality terms may weaken trade-secret arguments later.

Confidentiality duties and attorney-client privilege may be compromised
For lawyers, including patent practitioners, the use of AI tools can create ethical and professional-responsibility issues.  For example, USPTO rules require a practitioner to make reasonable efforts to prevent the inadvertent or unauthorized disclosure of, or unauthorized access to, information relating to the representation of a client.  The USPTO’s AI guidance specifically warns that using AI platforms for prior-art searching, application drafting, claim drafting, or Office-action responses may disclose client-sensitive or confidential information to third parties.

Privilege can also be complicated by the use of AI tools. For example, communications between a client and lawyer are normally protected when made for legal advice, but sharing the substance of those communications with an outside, public AI provider that is not covered by an appropriate confidentiality arrangement can potentially waive the attorney-client privilege or violate ethical rules relating to client confidentiality. Whether this occurs depends on jurisdiction, contract terms, and facts, but the risk remains.  If privileged information such as invention strategy, claim scope, prior-art analysis, or design-around discussions are submitted to a public AI tool, arguments could be raised that privilege was waived.

Public AI tools may retain, review, or use data in unexpected ways
Not all AI tools are the same – enterprise AI platforms and consumer-facing “chatbots” can have entirely different data handling policies. For example, some “enterprise” AI tools state that prompts and information shared are not used for model training by default, whereas personal or consumer-facing versions may use such information for AI training by default. Accordingly, care must be taken in selecting and researching an AI tool before it is used in a legal setting. 

Inventors should not assume that any friendly chat interface is confidential. Relevant questions include: Is there a contract with the provider? Is the provider bound to confidentiality? Is AI model training based on user input disabled? Are prompts retained by the provider? Who can review them? Are third-party integrations enabled? Are logs exported? Can workspace admins see chats? Are chats discoverable in litigation? Are data centers or reviewers outside the country? Without clear answers, submitting an invention disclosure is risky.

Export-control and foreign-filing-license problems may arise
Some technical information cannot simply be sent anywhere. The USPTO’s AI guidance warns that AI tools may use servers outside the United States, which can raise export-control, national-security, foreign-filing-license, or secrecy-order issues. It also notes that even servers located inside the United States can raise export issues if non-U.S. persons access controlled technology.

This is especially important for inventions involving sensitive fields such as aerospace, encryption, semiconductors, defense, nuclear technology, advanced materials, sensors, robotics, and biotechnology. A seemingly ordinary prompt could involve technical data that should not be exported or disclosed to foreign persons without review. The inventor may not know where the AI provider stores or processes the prompt data, who can access logs, or whether support staff is located abroad.

Inventorship and ownership may become harder to document
In the United States, an inventor named in a patent application must be a natural person.  Using an AI tool before filing does not automatically make the AI an “inventor.”  But heavy AI involvement can muddy the record of “who contributed what.” If an inventor gives a public AI tool a partial idea and the AI tool suggests a key technical feature that works its way into patent claims, a company may need to determine whether a human conceived the claimed invention, whether the AI merely assisted, and whether the resulting claim is properly supported by human inventive contribution.

Another practical issue is documentation. Patent ownership and inventorship disputes often turn on lab notebooks, invention disclosures, emails, prototype records, and dated drafts. AI chats can become part of that record. If they show that important claim elements came from an uncontrolled external system, a challenger may use them to raise questions about conception, derivation, enablement, or candor.

Contractual breaches may be triggered
Many inventions are developed under employment agreements, university policies, government grants, customer NDAs, joint-development agreements, sponsored-research contracts, or vendor agreements. Those contracts often restrict disclosure of technical information before patent filing or publication approval. Submitting the information to a public AI tool may breach those obligations even if patent rights survive.

This can matter in collaborations. For example, a company may not own all the information it “pastes” into an AI tool. It may include a customer’s confidential requirements, a supplier’s material specifications, government-funded data, open-source code, or a partner’s prototype results. The AI submission may therefore create not only patent risk, but also contract, compliance, and relationship risk.

Enforcement may be weakened
Even when a patent ultimately issues, pre-filing AI disclosures can become ammunition in litigation. Accused infringers may seek discovery of AI prompts, drafts, logs, and outputs. They may argue that the claims are anticipated by the inventor’s own earlier disclosure, that the patent is not enabled, that the inventors changed their story, that trade secrets were not kept secret, or that foreign rights were knowingly compromised.

The problem is not just whether the challenger wins. The existence of messy AI records can increase litigation cost, create embarrassing documents, and reduce settlement value. Patent enforcement depends heavily on a clean chronology of conception, diligence, filing, prosecution, commercialization, and the use of public AI can blur that chronology.

Practical Takeaways
These are only some of the issues to consider when using AI tools in the context of invention.  Given the complexities, what is a potentially safe path forward?

As in the past, to protect patent rights, the best approach is to file at least a provisional patent application before any public disclosure of the invention.  In the current environment, the safest policy is to treat the use of public AI tools as potentially amounting to such a disclosure.  Accordingly, an inventor should not submit technical details of an invention to a public AI tool prior to at least filing a provisional patent application.  

Another option is to use a vetted enterprise AI system under written terms that prevent sharing of confidential information, training of a publicly available AI model based on data entered into the system, access to prompts, and export of information outside the country.  This may ameliorate some of the dangers discussed above but requires careful attention to the details.

 The USPTO’s AI guidance does not prohibit AI use, but it emphasizes human governance, confidentiality, cybersecurity review, and responsibility for AI-assisted work.  To this end, a practical pre-filing AI policy accounting for known issues with shared information is a must. When AI assistance is needed, use sanitized prompts that omit the inventive concept, or ask general questions without revealing the actual invention.

In the end, protecting an invention involves turning private technical knowledge into a legally organized disclosure. Before that happens, public AI tools can accidentally move the invention into a problematic “gray zone” where patent protection might be compromised.

Reprinted with permission from the June 25, 2026 issue of The Legal Intelligencer ©2026 ALM Media Properties, LLC. Further duplication without permission is prohibited. All rights reserved.