Insight

Copyright, Patent, or Trade Secret Protection for AI Content: Challenges and Considerations

February 10, 2023

From creative endeavors like art and music to practical applications like translation and weather forecasting, artificial intelligence (AI) is producing more and more content and being used by inventors. Given the nuances in how AI technology is applied, those seeking to protect AI-generated content and inventions do not always have a simple or clear path forward under US and European laws.

United States

AI-generated content raises several questions when it comes to copyright and patent law, including whether the content is subject to copyright or patent protection and whether the content infringes on copyrights or patents.

Copyrights

As is more readily apparent after the launch of ChatGPT and DALL-E, AI is already creating content and inventions that would be protected by copyright in the United States if the authors/inventors were human. The US Copyright Office will register only original works of authorship created by a human, and determining how much a human was involved with the output of AI technologies is not always easy.

Some AI-generated content comes from training tools that use existing copyrighted works. US copyright law is still developing around the appropriate use of copyright-protected works as input for training AI applications.

Patents

AI technology can also create new products or processes with little or no human help. However, US patent law is reserved for inventions made by humans, so these AI-created products or processes cannot currently be patented. Unless US patent laws are modified to allow AI systems to be designated as “inventors,” which may take a long time, it is best to either use trade secret protection or make sure there is at least one human inventor involved in order to protect new products or processes created by AI technologies.

With respect to the patentability of AI inventions, it is often best to focus on something other than the AI technology itself in drafting patent claims, such as whether new data is being used in a novel way, or what technical problem was seen and solved. Patent claims should be closely tied to an application context for the underlying AI innovations; for instance, autonomous driving or image processing.

Patents do come with advantages: a lengthy monopoly, protection against reverse engineering and independent development, and the fact that they are easier to quantify for potential investors. However, the patent process is expensive, requires significant disclosure of technology, and can be difficult to obtain for AI innovations.

Trade Secrets

Trade secret protection is broader than patent protection and can essentially apply to anything with business value so long as it is kept secret, known only by those in the organization. Along with a low cost and no requirement to disclose technology, trade secret protection may be available for intellectual property that is not eligible for patent protection, such as raw data, training sets, and inventions made by AI technologies.

However, trade secret protection cannot stop reverse engineering and independent development, requires a lot of diligence and measures to maintain its confidentiality, and could be difficult to quantify for potential investors.

Considerations

When deciding whether to go the patent route or trade secret route, there are some general questions to consider:

  • Is there an invention? You cannot protect something with a patent if there is no invention. However, whether something is an invention is not always straightforward.
  • Will the invention be publicly visible? If people can see the invention, such as software on a phone or a phone chip, then patent protection is the only option.
  • Can the AI technology be reverse engineered or independently developed? If so, the invention should be protected by patent applications.

Europe

As in the United States, patenting an AI invention in the European Union would require inventiveness beyond the AI itself. The patent can also relate to modifying the AI and using the AI as part of another inventive process. The method in which an AI model is being trained, for example, could be patentable.

Similarly, autonomous AI systems “inventing” new products and systems are currently not patentable in the United Kingdom. Essentially there must be some human input in the innovation.

European Patent Office (EPO) Guidelines

AI inventions, addressed in Article 52 of the European Patent Convention, are examined in essentially the same way as inventions involving mathematical methods. For software patents, the EPO has a two-hurdle approach:

  • Patent eligibility, a relatively low bar where a single technical feature in the patent claim is enough to make it eligible
  • Inventive step, where only the technical features can establish nonobviousness in the patent claim

Keep in mind that even if a patent claim concerns a new and nonobvious use of AI, it could still be regarded as a mathematical method that lacks the technical features needed to support an inventive step. It is therefore important to include all technical specifications in the patent application, which requires significant work in talking with inventors and exploring other similar patent disclosures.

There are two safe harbor exemptions: technical implementation and technical application. Either exemption can be used and must be disclosed in the specification stage. However, it is worth noting that these exemptions can sacrifice the scope of protection afforded.

The EPO appears to require a description of what training data is suitable and/or at least one example set of suitable training data. The mere idea of applying machine learning to a known problem (with no further detail on how the model is used, implemented, or trained) is unlikely to lead to a patent.

Practical Tips for Disclosure

  • Describe potential technical implications. Is the design of the model framework motivated by internal functioning of the computer beyond mere programming?
  • Describe potential technical applications. Avoid options related to advertising, methods of doing business, financial applications, and games. Describe how the input, output, training, and/or model design serve a technical purpose, and avoid relying on subjective user preferences such as the positioning of an ad on a web page or the relevance of search results.
  • Sufficiency of disclosure is much more demanding in Europe compared to the United States. Strike a balance between disclosing enough information to meet EPO requirements and protecting commercially valuable information.

Whether seeking copyright, patent, or trade secret protection for AI-generated content or inventions, trusted legal counsel can help navigate complex rules and regulations to find the best path forward.

To learn more about AI and its impact on companies and industries, check out our Artificial Intelligence Boot Camp event series.