In 2023, the world-wide value of intangible assets (e.g., contractual rights and intellectual property (IP), such as patents, copyrights, and proprietary technology) reached USD57.3 trillion, surpassing pre-pandemic levels and representing an 8% increase from 2022.1 2 The United States was responsible for USD836 billion of this year-over-year increase.3 Major industry sectors—from pharmaceuticals to telecommunications—have intangible asset valuations that exceed the trillion dollar threshold.4
IP-heavy industries are increasingly turning to artificial intelligence (AI) to assist in the research and development of new products and services. For example, many innovative pharmaceutical companies are seeking to gain a competitive advantage by integrating AI into the laboratory development loop.5 As AI technology advances, it will increasingly drive corporate valuations of intangibles—potentially impacting trillions of dollars in corporate value. With such meaningful value at stake, it is important that board members and their advisors are up-to-date on the latest legal developments regarding protection of AI-related assets and ways to help mitigate risk of loss of value.
Uncertainties Exist with Respect to the Appropriate IP Strategy for Protecting AI and Its Output
Companies today are well acquainted with the use of patents and copyrights to protect their IP. These tools have historically afforded a company the ability to assess the scope of its IP protection as well as competitors’ patents and copyrights in order to understand potential barriers and vulnerabilities. Moreover, familiarity with patent and copyright protection has allowed companies to analyze their IP portfolio to determine the value such assets contribute to the company. The use of AI, however, complicates this landscape. There are uncertainties regarding obtaining patent and copyright protection on the components of an AI model and its output. This disrupts a company’s ability to understand the value of its AI-related assets and may lead more companies to consider relying upon other tools such as trade-secrets and contractual provisions to protect AI models and their output—which come with their own sets of draw-backs. Below, we describe the main avenues for protection of AI-related assets and key considerations for each.
Patents
A patent—a public document that includes a description of the invention and claims that define the scope of the patent’s protection—offers an excellent mechanism for creating value from a company’s intellectual property. The increased issuance of patents related to AI shows that patents will remain a key mechanism for protecting the value associated with AI inventions.6 There are, however, several issues that companies should consider when deciding whether a patent is the best vehicle to protect an AI model and its outputs. In the United States, uncertainty exists as to whether AI software, models, or platforms can qualify as patent-eligible subject matter.7 Should an applicant overcome the patent-eligible subject matter issue, uncertainties related to how an AI model is operating and will operate in the future may present issues for an applicant in describing and claiming the AI model. Should an applicant overcome these obstacles, the “black box” nature of a competitor’s AI model may present issues in assessing potential infringement by that competitor’s AI model, because it may be difficult to compare the patent claims to the competitor’s AI model to assess potential infringement.
The AI output or application is likely to avoid many of the patent hurdles often confronted by a patent application directed to an AI model, but it will have its own patent-related issues. Under the existing laws of several countries—including the United States and throughout Europe—AI cannot be listed as an inventor on a patent.8 If no human contributed to the invention, then a patent application is not viable unless these countries change their laws. It is unknown how these countries will treat applications that include inventive contributions from both humans and AI. Additional uncertainty exists as to how various countries will determine whether an invention generated by AI can show it was non-obvious. When addressing non-obviousness, the United States Patent and Trademark Office (USPTO) considers whether someone would be motivated to combine existing knowledge and would have a reasonable expectation that the combination would achieve the patented invention. This test is, in part, dictated by the finite capabilities of humans. The USPTO may need to rethink the obviousness standard, however, if AI can consider all knowledge and can predict the results of combining parts of that knowledge.9 These uncertainties present issues in how a company can assess the value of its IP portfolio and a competitor’s portfolio.
Copyrights
AI creates issues with copyrights too. The United States Copyright Office and courts in the United States have held that AI cannot be an author, rejecting, for example, contentions that the human prompting of the AI constitutes authorship.10 To the extent that aspects of an expression of an idea are attributable to AI, a company cannot protect those aspects of the work. Copyright issues also arise with the training of an AI model. To train a generative AI model, the model needs access to massive amounts of data. Some of this data may include copyrighted material, such as images or literary works. As of the date of this article, there is a host of pending litigations throughout the world alleging that the training of various generative AI models using copyrighted training data infringes those copyrights. It will take several years for the courts across various jurisdictions to resolve the arguments related to copyright infringement and the potential fair use of copyrighted material to train AI models.
Trade Secrets
Considering the ambiguities surrounding the legal protection of AI under current patent and copyright laws, companies may resort to trade secrets and contracts to define their rights and generate value. Trade secrets—intellectual property rights on confidential information by virtue of that information being commercially valuable because it is a secret, known only to a limited group of persons and for which the holder takes reasonable steps to keep confidential—avoid the hurdles presented in obtaining and enforcing a patent. However, having a trade secret with respect to an AI model would not prevent competition from independently developed or reverse-engineered AI models. To the extent that an AI model is public and a competitor can understand how it operates, trade secret protection may not provide much value. Considering the obligation to maintain secrecy to ensure the existence of trade secret protection, a company would have difficulty explaining the value that trade secrets provide the company—this might make it more difficult for the public and potential partners to understand the value of the company’s intangible assets.
Contractual Provisions
Contractual provisions offer the opportunity for companies to define the ownership and use rights that each has in the various components of the AI model and in the output from the AI model, along with each party’s obligations should infringement allegations arise. It is key for collaborators to address ownership, use rights, and indemnification responsibilities in such arrangement. But doing so only defines the relationship between the parties to the contract and does not offer protections against non-parties. Considering the benefits and drawbacks of the above IP protection mechanisms, it is imperative for a company to develop a strategy for how it will approach protection of its AI models and their outputs.
A Key to Navigating These Developments: Have an IP and AI Policy
There is not a one-size-fits-all approach to addressing IP issues that arise with respect to AI. Each situation will require thought as to the tailored approach appropriate for the product, the marketplace, and the targeted end user. Given that the various strategies to protect AI-related intangible assets each come with pros and cons, IP-heavy companies using AI would be well-served by board members ensuring that corporate leaders have weighed the various approaches and have in place a thoughtful protocol for considering the IP protection that can best protect the company’s AI and monetize its efforts.
A key approach to AI protection is implementing a robust AI policy identifying IP considerations associated with AI and outlining general preferences for the protection schemes. An appropriate AI policy should address a company’s approach to protecting its AI model, output, and procedures. The strategy to protect AI should address the various components of AI, including the hardware that implements the AI, the AI algorithm and model, and the output or application of the AI. New AI hardware and AI applications or outputs likely can meet the requirements for patent protection. A company may best protect its AI algorithm and model through trade secrets. The policy should explain the importance of taking certain steps so that a company will be in a position to protect its AI irrespective of the strategy it seeks to implement.
In order to ensure the ability to patent AI inventions, company policy should require the company to track the human involvement with the AI. This would include the persons responsible for the initial ideas as to the form and goals for the AI, developing the hardware for the AI, selecting the architecture of the AI, developing the algorithm, writing the source code, selecting the data to train, validate, and test the model, generating the prompts, interpreting the output of the AI, and using that output. Tracking human involvement will make it easier for a company to harvest potential inventions, identify persons that should be listed as inventors, and ensure that it has appropriate agreements that assign the rights from the individuals to the company.
To maintain AI as a trade secret, a company should establish that it has implemented reasonable procedures to maintain the information as a secret and develop evidence that shows the information derives independent economic value from not being generally well known. The AI policy should address confidentiality issues (including which employees may access information), place restrictions on third-party access, and require employees and third parties that have access to the AI to execute agreements in which they acknowledge the obligation to maintain the secrecy of the AI information.
The policy may address acceptable tools to use in generating the AI. For example, certain coding tools may use open source code that requires disclosure of modified code or other licensing obligations. If these tools are used, a company may unknowingly be encumbering its code with third-party rights or obligations. By exploring the obligations associated with these tools, a company can identify tools that will not result in unacceptable encumbrance and ensure that its employees avoid those tools.
The policy should require tracking of the origin of the data used to train, validate, and test the AI model. As evident from the numerous lawsuits filed by copyright owners, the potential for liability exists if data is used without obtaining rights to use that 11 As with coding tools, a company should identify sources of data that are acceptable to use with its AI models. The company should track what data is used with each version of its AI models so that it can delineate the rights and obligations associated with each version.
The policy should also establish practices that seek to avoid infringement of third-party IP. The procedures related to identifying acceptable coding tools and data sources will assist in avoiding infringement of third-party IP. The policy should establish a mechanism by which a company will respond to notices that allege a violation of a third-party IP right. The policy may also identify the situations under which it will conduct an IP landscape assessment. This assessment should allow the company to identify potential hurdles to the use of its AI model presented by third-party IP. It should also allow the company to identify whitespace where third-party IP does not exist. The company may exploit this whitespace by obtaining its own IP in that space.
The policy may address how a company will approach licensing third-party rights or licensing its AI. The strategy through which a company extracts value from its AI may dictate the approach it takes. An AI platform company will monetize its AI through third-party use of its AI models. It would be helpful to provide insight to employees concerning the general parameters under which it will allow access to the AI models and the parties’ rights in the model, data, and outputs. It should also consider what responsibility a party will maintain should an allegation of infringement of third-party IP arise. Other companies may monetize AI by improving internal processes or development of new products. Those companies will likely not engage in licensing of their AI models, as the companies’ exclusive right to use the AI model drives the value proposition.
These are key IP issues that a policy should address to provide clarity to the employees responsible for implementing a company’s AI strategy, but it is not an exhaustive list. Each industry and AI use case will present unique IP issues that a company must consider separate from these issues. It would be best for the company to address in its AI policy the IP issues unique to its industry or AI use case.
_______________________________
1 Brand Finance Global Intangible Finance Tracker 2023, at 16.
2 Id. at 10.
3 Id. at 22.
4 Id. at 19.
5 Andrew Dunn, Q&A: Aviv Regev talks three years at Genentech, future of AI in biotech, at 2-4, Endpoints News (Dec. 11, 2023).
6 In 2017, the United States Patent and Trademark Office issued 3,267 AI-related patents. That number increased to 18,753 in 2021. Rose Acoraci Zeck, Analysis: Patents Forecast Widespread Reach of AI Tech in 2023, Bloomberg Law (Nov. 13, 2022).
7 2019 Revised Patent Subject Matter Eligibility Guidance, Federal Register (Jan. 7, 2019).
8 The following countries are among those that explicitly denied inventorship to AI: the United States (Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022)); the United Kingdom (Thaler v Comptroller-General of Patents, Designs and Trademarks [2023] UKSC 49); South Korea (Seoul Administrative Court [Seoul Admin. Ct.], 2022GuHap89524, May 12, 2023); Australia (Thaler v Commissioner of Patents [2021] FCA 879.); and the members of the European Patent Office (register publication with grounds for denial of a patent listing the inventor as an AI system).
9 President Biden’s October 30, 2023 Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence directs the USPTO to issue guidance that addresses the interaction of AI and IP by July 26, 2024. See Section 5.2(c) of Exec. Order No. 14110, 88 F.R. 75191 (2023). This guidance will likely address issues that the use of AI presents with the obviousness assessment.
10 Thaler v. Perlmutter, No. 22-CV-01564-BAH (D.D.C. Aug. 18, 2023).
11 Isaiah Poritz, OpenAI Faces Existential Threat in New York Times Copyright Suit, Bloomberg Law (Dec. 29, 2023).
