Protecting AI-Generated Inventions as Trade Secrets Requires Protecting the Generative AI as Well

“If the generative AI is not itself protected as a trade secret, then it may be vulnerable to, for example, copying or reverse engineering…[and] an invention or other information generated by the AI may be readily ascertainable through proper means and thus not be a trade secret.”, technology, business, and academic professionals currently are debating whether an invention autonomously generated by artificial intelligence (AI) should be patentable in the United States and elsewhere. Some proponents of patentability argue that if AI, by itself, is not recognized as an inventor, then AI owners will lack protection for AI-generated inventions and AI innovation, commercialization, and investment (collectively, AI innovation) will be inhibited as a result. Some of those proponents further argue that, without patent protection as an option, AI owners increasingly will opt for trade secret protection, which by design reduces public disclosure of corresponding inventions and, as such, still will inhibit AI innovation. Some opponents of patentability, on the other hand, argue that patenting AI-generated inventions will promote those inventions and discourage human-generated inventions, thereby reducing human innovation and ultimately competition, because patent ownership will become concentrated, or more concentrated, in fewer entities—in particular, large, well-funded entities.

As that debate continues, innovators are left to deal with reality. The U.S. Patent Act currently does not offer protection for AI-generated inventions where AI is the only named inventor. Because of that void, trade secret protection for AI-generated inventions may become more popular. Indeed, such an outcome is likely because the Defend Trade Secrets Act (DTSA), for example, can protect inventions and other information autonomously generated by AI.

Eligibility for trade secret status under the DTSA is one thing. Satisfying the DTSA’s requirements for trade secret status is another thing entirely. More specifically, the requirements that an owner take reasonable measures to keep its information secret and that the information not be readily ascertainable through proper means present a unique challenge where the information—i.e., the potential trade secret—is AI-generated information.  In short, to satisfy those requirements, an owner must not only take reasonable measures to maintain the secrecy of the AI-generated information but also the secrecy of the underlying, or generative, AI such that neither the AI-generated invention nor the generative AI is readily ascertainable through proper means.

Ultimately, satisfying the DTSA’s requirements for trade secret protection is achievable through a focused intellectual property (IP) strategy. Such a strategy may include dual trade secret protection where, as noted above, both the AI-generated information, such as an AI-generated invention, and the generative AI are protected as trade secrets. Alternatively, such a strategy may include complementary IP protection where the AI-generated information is protected as a trade secret and the AI itself is protected partially by a patent and partially as a trade secret. In either case, concerns about increased trade secret protection inhibiting AI innovation appear to be overstated.

AI-Generated Inventions Currently Are Not Patentable in The United States

As discussed in prior articles, in 2021, the U.S. District Court for the Eastern District of Virginia addressed in Thaler v Hirshfeld, 558 F. Supp. 3d. 238  (E.D. Va. 2021), whether “an artificial intelligence machine can be an ‘inventor’ under the Patent Act,” and held that it cannot be. So, at least for now, where an invention is generated by AI, and a corresponding patent application names only AI as an inventor, the invention is not patentable in the United States.

Thaler has appealed the district court’s decision to the U.S. Court of Appeals for the Federal Circuit.  The Federal Circuit held oral arguments on June 6, 2022 but has not yet rendered a decision.

If the Thaler holding remains the law, then the decrease in innovation that proponents of patentability predict may come to fruition. More likely, however, is that owners of AI-generated inventions will seek other means of protection for those inventions, such as trade secret protection under the DTSA, and proponents of affording inventor status to AI will ostensibly lobby Congress to amend the Patent Act.

AI-Generated Inventions Can Be Trade Secrets

An AI-generated invention is eligible to be a trade secret because any information is eligible to be a trade secret. 18 U.S.C. § 1839(3) (defining “trade secret” broadly as all forms and types of financial, business, scientific, technical, economic, or engineering information, including patterns, plans, compilations, program devices, formulas, designs, prototypes, methods, techniques, processes, procedures, programs, or codes, whether tangible or intangible, and whether or how stored, compiled, or memorialized physically, electronically, graphically, photographically, or in writing”). Indeed, there is no requirement under the DTSA that an inventor, developer, or generator of a trade secret be identified or be a natural person.

The absence of an identification requirement makes sense because “[n]ovelty, in the patent law sense, is not required for a trade secret . . . .”Kewanee Oil Co. v. Bicron Corp., 416 U.S. 470, 476 (1974) (internal citation omitted). In other words, multiple persons can independently develop or derive the same trade secret, and all of them are entitled to the protections of the DTSA provided they comply with the requirements thereof. See 18 U.S.C. § 1839(6)(B).

The absence of a natural person requirement demonstrates the DTSA’s neutrality on who or what invents, develops, or generates information. That neutrality contrasts with the DTSA’s requirements for the character and treatment of information. That is, to be a trade secret, information must derive “independent economic value, actual or potential, from not being generally known to, and not being readily ascertainable through proper means by, another person who can obtain economic value from the disclosure or use of the information” and “the owner thereof [must have] taken reasonable measures to keep such information secret . . . .” 18 U.S.C. § 1839(3). Of course, if those requirements are satisfied, then the owner or, as noted below, a licensee of the trade secret may seek injunctive relief, damages, or both for actual or threatened misappropriation. 18 U.S.C. § 1836(b)(3).

The DTSA defines the “owner” of a trade secret to be “the person or entity in whom or in which rightful legal or equitable title to, or license in, the trade secret is reposed[.]” 18 U.S.C. § 1839(4).  Thus, where the trade secret is an AI-generated invention, the owner of the generative AI is the owner of the trade secret. Ownership of the AI likely would be established through (1) a contract, such as a proprietary information or invention agreement or other assignment; (2) a relationship, such as employer-employee relationship; or (3) a combination of both a contract and a relationship.

In that regard, AI and its output are treated no differently than any other production means, such as a machine or software, and its output. As a simple example, if Company X owns a machine that manufactures widgets, then Company X would own the widgets, too, even though the machine manufactured them. The same result ensues where Company X owns software created by its employee and the software generates output. So, if Company X owns the means, such as AI, a machine or software, then Company X owns the corresponding output and can enforce its rights and interests therein. Sensibly, the sophistication of the means—whether AI, a machine, software, or other tool—does not affect ownership of the output.

As noted above, one of the requirements for trade secret status is the owner of the information at issue must take “reasonable measures to keep such information secret[.]” 18 U.S.C. § 1839(3)(A).  In practice, that means the information is the object of affirmative, ongoing protective steps. But what does that mean when the invention—i.e., the potential trade secret—was autonomously generated by AI?

Reasonable protective measures for an AI-generated invention can include, for example: storing such information in a secure location with password protection and dual-factor authentication; providing access to the information on a need-to-know basis; implementing confidentiality policies and procedures; requiring employees, vendors, and business partners to execute confidentiality agreements; and electronically monitoring access to and other activity involving the information. Such measures are not unique to AI-generated inventions and often are taken to protect various trade secrets, regardless of how or by whom or what the trade secret is generated. Moreover, such measures can be supplemented or enhanced based on the specific information and circumstances at issue.

Reasonable protective measures for only the AI-generated invention may not be enough to establish or maintain its status as a trade secret, however. More specifically, if the generative AI is not itself protected as a trade secret, then it may be vulnerable to, for example, copying or reverse engineering. If the AI is copied or reverse engineered, then an invention or other information generated by the AI may be readily ascertainable through proper means and thus not be a trade secret.

Those vulnerabilities demonstrate that reasonable measures to protect the secrecy of the generative AI or certain parts of it also should be taken. Depending on the nature of the AI, protective measures like those set forth above could maintain the secrecy of the AI or certain parts of it, such as inputs, input processing, and algorithms applied to inputs. While some of that information, such as an input, may not by itself be a trade secret because it is publicly or generally known or available, a combination of elements, parts, or steps, such as a process, formula, or recipe, can be a trade secret. See, e.g., Airfacts, Inc. v. De Amezaga, 909 F.3d. 84, 96 (4th Cir. 2018) (“a trade secret can exist in a combination of characteristics and components, each of which, by itself, is in the public domain, but the unified process, design and operation of which, in unique combination, affords a competitive advantage and is a protectable secret”) (internal citation omitted; collecting cases).

The timing of protective measures for the generative AI also matters. For example, if it already is publicly disclosed or can be copied or reverse engineered from goods or services available in the marketplace, then protective measures for the AI would be, and protective measures for the AI-generated invention may be, unavailing. That is, the AI would be, and the AI-generated invention may be, readily ascertainable through proper means and, thus, trade secret status for the AI and AI-generated invention may be non-existent or fleeting.

There also may be an important, patent-related consideration in the above-described scenario. If, for example, the AI is disclosed in a patent or published patent application, then the invention that AI generated may be readily ascertainable from the patent or application. But, if such ascertainment results from patent infringement—i.e., practicing a valid, enforceable patent claim without authority—then liability for patent infringement and trade secret misappropriation may attach. See 35 U.S.C. §§§ 101-103, 112 and 271; 18 U.S.C. § 1839(3), (5). In other words, patent infringement is not “discovery by fair and honest means” or “proper means.” 18 U.S.C. § 1839(3), (5); see Kewanee Oil, 416 U.S. at 476 (“A trade secret law, however, does not offer protection against discovery by fair and honest means, such as by independent invention, accidental disclosure, or by so-called reverse engineering, that is by starting with the known product and working backward to divine the process which aided in its development or manufacture.”) (internal footnote omitted).

The Concern About a Shift to Trade Secret Protection May Be Overstated

The concern that a shift to trade secret protection will inhibit AI innovation may be overstated for three reasons. First, trade secret law protects and, therefore, encourages innovation. Kewanee Oil, 416 U.S. at 481 (“The maintenance of standards of commercial ethics and the encouragement of invention are the broadly stated policies behind trade secret law.”).

Second, while an invention protected as a trade secret will not be publicly disclosed, “[t]rade secret law promotes the sharing of knowledge, and the efficient operation of industry; it permits the individual inventor to reap the rewards of his labor by contracting with a company large enough to develop and exploit it.” Id., at 493. Indeed, such sharing, with appropriate persons and under appropriate circumstances, is an inherent part of successfully exploiting a trade secret in a commercial enterprise. See, e.g., id., at 475 (“This necessary element of secrecy is not lost, however, if the holder of the trade secret reveals the trade secret to another ‘in confidence, and under an implied obligation not to use or disclose it.’ These others may include those of the holder’s ‘employees to whom it is necessary to confide it, in order to apply it to the uses for which it is intended.’”) (internal citations omitted).

Third, capitalism—i.e., private industry—along with universities and other centers of research and development, drive innovation. Id., at 491. That is, “if the invention, though still a secret, is put into public use, the competition is alerted to the existence of the inventor’s solution to the problem and may be encouraged to make an extra effort to independently find the solution thus known to be possible.” Id.

The Trade Secret Status of an AI-Generated Invention Can Depend on Whether and How the Generative AI Is Protected

In sum, there are three primary choices for protecting AI: patent protection, trade secret protection, or a combination of both. Copyright protection may be available for certain aspects of the AI, such as software. The type of protection chosen can substantially impact the trade secret status of an AI-generated invention.

If the generative AI is disclosed in a patent, published patent application, or available prosecution document, then a person may use the patent, application, or document to readily ascertain, or recreate, the AI-generated invention. While the person may be liable for such use, a substantial, practical consequence is the trade secret may be lost—i.e., the AI-generated invention may be publicly disclosed. Kewanee Oil, 416 U.S. at 484 (“By definition a trade secret has not been placed in the public domain”); Kinship Partners, Inc. v. Embark Veterinary, Inc., 3:21-cv-01631-HZ, at *17 (D. Or. Jan. 3, 2022) (“A trade secret once lost is, of course, lost forever.”) (internal citation omitted).

If the AI is protected as a trade secret, then there is an interdependent obligation on a trade secret owner: reasonably protect both the generative AI and the AI-generated invention as trade secrets.  In other words, if reasonable measures to maintain the secrecy of the AI are not taken, and, on top of that, the AI can be copied or reverse engineered, then there may be a domino effect. The generative AI will lose its trade secret status (simply because reasonable protective measures were not taken) and the AI-generated invention may lose its trade secret status because it may be readily ascertainable through proper means—i.e., able to be generated from the copied or reverse engineered AI.

Finally, if the AI is protected partially by a patent and partially as a trade secret, then the patent, including any underlying application and its prosecution, may not be means to readily ascertain the AI-generated invention. That is, the patent will not be a “readily ascertainable roadmap” where it discloses what is necessary to satisfy patentability requirements but does not disclose the part of the AI protected as a trade secret. Robust IP protection for the AI and AI-generated invention therefore may be best obtained through a combination of patent and trade secret protection.


Image Source: Deposit Photos
Image ID:222059474


Warning & Disclaimer: The pages, articles and comments on do not constitute legal advice, nor do they create any attorney-client relationship. The articles published express the personal opinion and views of the author as of the time of publication and should not be attributed to the author’s employer, clients or the sponsors of

Join the Discussion

4 comments so far.

  • [Avatar for Anon]
    July 26, 2022 10:02 am

    Thanks Anon2 – I recognize the point that you think the authors are merely stating that to enact Trade Secret protection, that “generation of” would also need be treated and protected as Trade Secret.

    This is a false statement though.

    One of the “troubling” aspects in the computing arts related to AI (and present at the precursor levels) is that “traceability” of end product to inputs (the “how did the machine arrive there”) is often lacking.

    I have personal knowledge of several innovations (unpublished at this point) addressing this – but the larger point remains: even as the level of what needs to be protected under Trade Secret treatment is not as the authors portray, Trade Secrets simply is not the most desirable path for providing innovation protection.

    Further note, as Mr. White’s comment was not visible to me when I posted: true AI removes the possibility that Mr. White advances, as the humans that program, pose, or even validate/select cannot be listed as inventors as they fail to meet the legal definition of inventors. I have provided very easy to follow thought experiments explicating this point (the presentation of a black box to a non-inventor). Sure, clients may opt for this “fix,” but if they do so, they run the high risk of losing protection only after a patent may finally be achieved during an enforcement attempt. AS IS, the state of innovation provides that certain innovations (with AI as inventor – either solo as in DABUS, or more perniciously, as part of a team) are simply NOT protectable under patents.

  • [Avatar for Anon2]
    July 25, 2022 09:23 am

    I don’t view this article as “trying to weaken the perception that Trade Secret may be a viable path of protection” at all. The article merely posits that to protect AI-generated inventions, you must also protect the AI that generated it, or you could lose protection of both. That would apply regardless of the industry.

  • [Avatar for Anon]
    July 25, 2022 08:11 am

    I do not agree with the main assertion of this article, as it both argues too little and too much.

    Bottom line is that what they seem to identify as a ‘weakness’ is true of ALL items to which one would wish to try to protect by trade secret — at the same time that they appear to be trying to weaken the perception that Trade Secret may be a viable path of protection.

    My stance on Trade Secrets has always been one that they are ‘the enemy’ of patent protection – and are inimical to what the concept OF patent protection is meant to present. This is a point that the authors and I would appear to agree to, BUT – that being said – I still think that Trade Secret protection IS something on the table for my clients, and thus, should not be ‘cheapened’ in an incorrect manner as is implicated here.

    A further note – this is the first time that I have seen the notion (being submitted almost as a casual nod) that allowing patents for AI would not be good for human inventors because of some “consolidation by Big Money” effect. I would be appreciative if the authors could provide some references for this line of view (I am not necessarily disagreeing with this line of view, and have not drawn my own conclusions on such).

    Lastly, as one of the authors appears to be related to FinTech (Mr. Pelletier), I am surprised that no mention is made of the competing non-patent law (finance disclosure law) that is driving requirements in traceability in use of machine learning (a pre-cursor computing art to true AI) as it applies to financial flows.

  • [Avatar for George White]
    George White
    July 24, 2022 06:01 pm

    You state, correctly, an invention where only an AI program is listed as the inventor can not be patented in the U.S. If humans who programed the AI program, posed the question to the AI program or validated and selected among solutions suggested by the program are listed then the innovation can be patented.

    This seems more straightforward than trying to use trade secret protection but is not addressed in your piece.