The Situation: Artificial intelligence (“AI”) technology is exploding across virtually all industries. Technology companies are innovating at warp speed, and even companies that do not principally identify as “technology companies” are becoming increasingly “high tech” in how they deliver goods and services. Innovation is outpacing the law—many aspects of AI legal protection are still open questions.
The Result: With innovation comes imitation … and infringement and misappropriation. Companies must be vigilant in protecting and enforcing their intellectual property (“IP”) rights in AI.
Looking Ahead: The time is now to devise a legal strategy to protect IP in AI. Such strategy should consider not only patent protection but also copyrights and trade secrets.
The term “artificial intelligence” was coined more than 60 years ago and since has been the subject of countless science fiction books and movies. Just a few months ago, Blade Runner 2049 hit the theaters as a sequel to the original 1982 Blade Runner film, which imagined a futuristic world—in the year 2019—where synthetic, bioengineered humans run rampant. Here we are in 2018, and while Blade Runner‘s dystopian vision has thankfully not come to pass, there is no doubt that the future is now: AI has begun to infiltrate virtually all business sectors and nearly every aspect of the consumer experience.
The legal implications for AI are complex and constantly evolving, and they often take a back seat to the pressing business demands of bringing innovative technologies to market. That said, it is important for businesses today to implement a proactive and comprehensive strategy to protect their IP rights in AI. Such strategy should consider not only patent coverage but also copyright and trade secret protection, particularly for aspects of the IP that may not (rightly or wrongly) be patent eligible.
Patents can be a powerful form of protection for AI-related IP, particularly because independent creation is not a defense to patent infringement. But patents are not a complete solution. As an initial matter, patents are unavailable for certain facets of AI. For example, patents do not protect data compilations, such as AI training sets, a programmer’s particular expression of source code, or other types of proprietary information that may be competitively advantageous and constitute a trade secret.
Moreover, the patent system in the United States presents uncertainties regarding patentability as it pertains to certain aspects of AI, such as software—since the Supreme Court’s decision in Alice, many of these patents have been attacked and invalidated as patent ineligible under 35 U.S.C. § 101. Under this regime, the decision to seek a patent may have dire consequences—if a patent is sought but not obtained, or granted and then invalidated, the subject matter may have become public, rendering not only patent protection but also trade secret protection unavailable.
Finally, patents have a limited term of protection (20 years from the earliest filing date), after which the subject matter becomes dedicated to the public. In contrast, a copyright term is much longer (the author’s life plus 70 years), and trade secret protection can theoretically last forever, as long as secrecy is maintained and the information is not publicly known. Thus, trade secret and copyright protection should be taken into account when developing an effective AI protection strategy.
AI and Trade Secrets
Trade Secret Protection
Federal and state trade secret laws protect economically valuable secrets. Protectable information includes formulae, compilations, programs, methods, techniques, processes, designs, and codes. Unlike a patent, no application or registration is required to obtain trade secret protection; rather, trade secret protection arises automatically provided that the trade secret owner can demonstrate that the information creates a competitive advantage by virtue of its secrecy and that reasonable measures have been taken to maintain its secrecy. Many AI system elements are well-suited for trade secret protection, such as: neural networks, including modular network structure and individual modules; training sets, data output, and other data; software including underlying AI code and AI-generated code; and learning, backpropagation, and other algorithms.
“Reasonable measures” to maintain the secrecy of a trade secret may vary depending on a company’s size and resources, but they can include physical and technical solutions to limit and monitor access to trade secret information. Physical solutions include locked cabinets and server rooms, whereas technical solutions may involve multi-factor authentication, mobile device management, and data loss prevention software. Written policies should dictate trade secret management, and employees who have access to trade secrets should be limited in number and contractually required to protect company confidential and trade secret information from improper disclosure. When sharing trade secrets with business partners, nondisclosure agreements should have secrecy obligations, audit rights, and provisions for post-relationship control. Object code given to customers should include digital rights management and robust licensing terms containing anti-reverse engineering provisions.
Developing a Strategy
As a preliminary matter, companies should consider a trade secret program that goes beyond “reasonable measures.” A robust “belt-and-suspenders” is advisable, because whether the measures were in fact “reasonable” is a common battleground in litigation. But more importantly, robust measures at the outset will diminish the likelihood of misappropriation and enable rapid detection of misappropriation when it occurs.
To this end, organizations should consider undertaking a process to identify and categorize their trade secrets. Some companies find it useful to understand their most valuable trade secrets, and to develop a trade secret program that prioritizes trade secret protection resources around their “crown jewels.” This requires discipline across the organization, as AI and associated software development is an ongoing, evolutionary process involving multiple people and many revisions; trade secret identification may get lost in the shuffle. All too often when litigation arises, there is confusion surrounding what the exact trade secrets are and why a competitive advantage is gained by virtue of their secrecy. If a particular AI trade secret is vital to the company—for example, a large training dataset—then significant resources and care should be devoted to its protection.
AI and Copyright
Software Copyright Protection
Although we may not think of software as a literary work, it is protected as such under U.S. copyright law. Copyright protection in software extends to all of the original expression embodied in the software, but not to its functional aspects, such as algorithms, formatting, logic, or system design. The functionality of AI software is often precisely what makes it valuable, so litigants must exercise care to connect that functionality (which is not covered by copyright) to the specific original expression in the software (which is).
Issues that arise in copyright protection of AI include:
Authorship. Following the infamous “Monkey-Selfie” case, the U.S. Copyright Office clarified that “[t]o qualify as a work of ‘authorship’ a work must be created by a human being.” This creates challenges for the copyrightability of AI-generated works.
There are also hurdles to clear with human-generated software. The work for hire doctrine in the current copyright law does not extend to commissioned works of computer software. Accordingly, agreements with third-party software developers must include express assignment language as well as work for hire language to specifically establish that any copyrighted software resulting from the project will be owned by the company and not the developer. In addition, there are challenges relating to third-party code and open-source code that must be evaluated during the software development process.
Copyright Registration. Although not required to secure a copyright, registration confers important benefits. Registration is a prerequisite to bringing an infringement suit; statutory damages and attorneys’ fees are unavailable absent timely registration; and timely registration constitutes prima facie evidence of a valid copyright, shifting the burden of proof on that issue in an infringement action. Note that revised software may contain new copyrighted material requiring a new registration.
Copyrighting AI Data. Pure data is not copyrightable, but data compilations can be copyrightable—particularly when raw data has been manipulated and organized into structured datasets. To the extent available, trade secret protection is generally preferable, however, for AI datasets.
Redaction of Trade Secrets in Copyrightable Code. Where copyrighted software includes trade secrets, they should be redacted from the code deposited with the Copyright Office, as the deposit is a public record. In addition, a letter should be submitted to advise that the software contains trade secrets.
Developing a Strategy for Protecting Copyrighted AI Software
The adage “an ounce of prevention is worth a pound of cure” readily applies to software copyright registration. When software theft occurs, organizations are all too often delayed in bringing suit because software is not registered; even when registered, challenges arise in litigation because there are difficulties in identifying the specific version(s) that have been infringed. Organizations should thus develop a strategy for timely registering copyrightable software. Because software is often undergoing nearly continuous updates, it may be impracticable to register every version—but, at a minimum, the company should have some type of protocol in place where newly developed software versions are registered (e.g., for critical updates, once a quarter, every time roughly 20 percent of the code changes, etc.). “Version control” is a critical component of copyrighted software protection strategy because, in litigation, the full code for the “copyrighted work” will likely need to be produced and will need to match the version that is registered and the version that the defendant copied. Further, robust licenses, including end user license agreements, should control use of the licensed software and allocation of ownership of code and/or data developed from the licensed software.
AI today is less about imagined scenarios in the far-distant future and more about real-world applications happening right now or on the horizon. Organizations having a robust and well-thought-out AI protection strategy in place will be best positioned to enforce their IP rights when infringement and misappropriation occur and, better yet, to prevent unlawful conduct before it occurs. Trade secrets and copyright provide unique protection for IP that cannot be achieved by patents alone.
Three Key Takeaways
- A number of AI system elements are good fits for protection as trade secrets.
- The more valuable a particular AI trade secret is to a company, the more resources should be applied to its protection.
- Copyrights can be an important part of an AI protection strategy, but they raise certain issues relative to functionality, authorship, registration, and other matters.
The original press release can be found here on Jones Day’s website.