Can AI keep a secret?

The field of artificial intelligence has accelerated exponentially in recent years. Various AI systems are now available for commercial use, with several - such as ChatGPT - open to the general public at low or no cost. AI is already capable of generating content - such as chatbot messages and "deep fakes" - that can be difficult if not impossible to distinguish from human works. As AI systems continue to grow in sophistication, many believe they will soon approach (or perhaps exceed) humans in their ability to conceive and develop new technological inventions.

AI already plays a role as an important tool for technological innovation in a wide range of industries. The pharmaceutical industry, for example, uses AI to help identify and develop new drugs. As AI continues to transition from being a mere assistive tool to being independently capable of innovation, those wielding AI must give careful thought to how AI-generated inventions can best be protected under existing intellectual property paradigms.

Intellectual property law offers human inventors and creators three primary vehicles for protection: copyrights, patents and trade secrets. The story is different for AI-generated creations. Notwithstanding that copyright protection is generally unavailable for technological inventions, most jurisdictions will not grant a copyright to AI-generated works lacking significant human contribution.

Patent protection is similarly unavailable in most jurisdictions for inventions conceived solely by an AI "inventor," with South Africa being an exception. In the United States, for example, an "inventor" under the Patent Act must be a human being. See Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022). There is, however, a debate in many jurisdictions whether patent protection is available to inventions jointly conceived by humans and AI. It is also untested whether the owner of an AI machine can be substituted as the named inventor for purposes of obtaining a patent. Notably, the European Patent Office has suggested that this may be permissible for European patents.

In contrast, trade secret protection is agnostic to whether a human or AI created an invention. It is thus a potentially attractive option for those seeking to protect and monetize inventions created by their AI systems - especially in jurisdictions where patent protection is unavailable or uncertain. But a fundamental prerequisite for trade secret protection is that the technology be kept secret. This presents a difficult choice for AI owners and operators seeking global protection - a patent disclosure in one jurisdiction undermines trade secret protection in all others.

The secrecy requirement also raises the related and difficult question of whether AI can, in fact, keep a secret. Those that seek trade secret protection for AI-generated inventions must carefully consider the technological and contractual limitations of their AI system against trade secret eligibility requirements. These requirements include, at least in the United States, engaging in "reasonable measures" to ensure the technology's secrecy, among others.

On the technological front, careful attention should be paid at the outset to how your AI system operates, including how the AI generates content and where that generated content is stored. At a simplistic level, AI works by taking input data from the user and generating an output. Those who may ultimately want to rely on trade secret protection for the AI's output results should examine what protections are in place to control how and whether that output is stored, who may access it, and whether the output may be used in subsequent queries by the AI. Even in a non-public AI system proprietary to a single company, the AI-generated output may be too readily accessible to others such that "reasonable measures" have not been taken to ensure the secrecy of the information.

The increasing prevalence of public AI tools controlled by third parties further exacerbates the technical challenge of keeping AI-generated inventions secret. Many AI tools, such as ChatGPT and its ilk, are cloud-based and can be used through a web interface or API. Companies employing public AI products in their workflow may find it even more difficult to adequately protect trade-secret level information if they do not have adequate control over how and where their AI-generated content is stored, and who has access to the information.

With third-party-provided AI systems (whether proprietary AI acquired for in-house use, or publicly available AI tools), it is also important to analyze relevant contractual agreements and terms-of-use for their potential impact on trade secret eligibility. As a first step, an AI user should ensure that they have ownership of the AI-generated content, and that the content will be maintained as confidential by the AI provider. Securing ownership rights and confidentiality may be particularly challenging for publicly available AI tools. For example, the terms-of-use for some providers expressly grant ownership of AI-generated content to the AI provider, destroying any possibility for the user to obtain trade secret protection over the AI-generated content. Other AI providers grant ownership of AI-generated content to the AI user, but may retain rights to use the data input into and/or output from the AI for certain purposes, such as to develop or improve the provider's AI models. This data license to the AI provider can have ramifications on trade secret eligibility, depending on exactly what data is accessible to the AI provider, how it is stored and used by the AI provider, and what confidentiality obligations are in place.

As AI use continues to accelerate, those using AI systems to innovate may find out too late that there are few to no intellectual property protections available for their AI-generated inventions. With patent protection for AI innovations currently uncertain, if not entirely unavailable in the United States and elsewhere, trade secrets offer a possible mechanism for protection. But stake-holders seeking to leverage trade secret protection would do well to pay early attention to the technological underpinnings of their AI systems and their contractual agreements with AI providers to ensure that trade secret prerequisites are met.

James L. Davis Jr. is a partner in the Intellectual Property Litigation Group of Ropes & Gray LLP and is resident in the Silicon Valley Office.

Scott Taylor is a senior attorney in the Intellectual Property Litigation Group of Ropes & Gray LLP and is resident in the Boston Office.