A Legal 360 for Intellectual Property for AI

This article reflects only the present personal considerations, opinions, and/or views of the authors, which should not be attributed to any of the authors’ current or prior law firm(s) or former or present clients.

All major industries are in the process of, or are considering, adopting Artificial Intelligence (AI). Whether it is in their organization, logistics, or products, companies are experiencing tectonic shifts in the commercial landscape. Commensurate with these tectonic shifts are new well-springs of intellectual property. And all points in the AI pipeline are affected: data collection and curation, model creation and training, and the output of AI systems.

In this article we survey the existing intellectual property legal protections and practice tips for protecting AI-related technology. Copyright, patents, and trademarks, and trade secretes can all be use together to protect different aspects of AI. Further privacy is an area of ongoing importance, and specifically the data acquisition and curation. While the following discussion tends to focus on the United States, some observations may apply in other countries.


Copyright protects original works of authorship including literary, dramatic, musical, and artistic works and encompasses a wide variety of works including poetry, novels, movies, songs, computer software, and architecture. Today, copyright does not protect works that are created by non-humans. Just like a picture taken by a monkey cannot be copyrighted, neither can the stories, songs, images, source code and the like that are created by AI. At the same time, copyright does afford protection to authors for works that are later generated by AI. If AI generates a work that was previously copyrighted by a human author, the owner of the AI model or the company using the AI can be on the hook for copyright infringement.

The heavy data requirements for training AI can also result in copyright infringement. The data used to train AI is often obtained by scrubbing the Internet and may include someone’s copyrighted data, leading to copyright infringement. One defense to copyright infringement is the fair use doctrine.1 Fair use is a defense to infringement that allows the unlicensed use of copyrighted material under certain conditions. The law of fair use is complicated and heavily fact-dependent. There are several litigations in U.S. that question whether fair use doctrine can be used as a defense to train AI.2

Congress is also looking at AI-related copyright issues. On July 12, 2023, the Senate Judiciary Subcommittee hearing on AI opened with an AI-generated version of Frank Sinatra’s song “New York, New York.”3 The song was sung in Frank Sinatra’s voice with lyrics about regulating AI and served as an example that today this song cannot be protected by copyright, yet can be used as a basis for copyright infringement.

Stakeholders in AI should be cognizant of the legal landscape that AI cannot be the author of copyrightable work. In addition, companies engaged in data acquisition, AI training, and generative AI should be cognizant of the data they use and where the data comes from before being sued for copyright infringement. While the fair use doctrine may be available as a defense, the inquiry is highly factual, expensive, and time consuming. Further there is no definitive case law to say that it may be successful.


Patents are a right to exclude others from making, using, offering for sale, selling, and importing the patented invention.4  Patents are a legally sanctioned monopoly that exists for 20 years in exchange for disclosing how to make the invention to the public. Patents offer some of the strongest protection for intellectual property and can be used both offensively—through patent enforcement—and defensively—by preventing other lawsuits. Patents are often part of the calculus to obtain funding for start-ups. But there are challenges to obtaining and enforcing an AI patent. As AI is software and often improves a way of doing business, a question exists whether AI falls within the judicially created exceptions for patent eligible subject matter. This is a grey area as there is no definitive case law that AI inventions are not patent eligible. The United States Patent & Trademark Office has released guidance that suggests it views patenting neural networks, the building blocks of AI, as permissible.5  Similar subject matter eligibility challenges exist in other jurisdictions, including the European Union.6

Stakeholders in AI should be cognizant of the legal uncertainty surrounding the eligibility of AI systems for patenting. At the same time, AI is a rapidly evolving field, and companies are well served in obtaining patent protections for their AI technology to gain a competitive advantage. Thus, attorneys savvy in both AI procurement and litigation are often used to handle AI related patent issues.


Trademarks allows the public identify goods and services bearing the mark with their source. In that vein, trademarks may be used to protect the brand of the new AI products and services for companies creating and employing AI. For a field growing more and more crowded, a well-designed and distinctive trademark may help a product or service move ahead of the pack.

Stakeholders in AI will need to consider the effectiveness of their trademarks in a field where there is a stream of new entrants. Whether there were missteps with an initial AI product offering or the product is well-received by the market, distinguishing from past missteps or the competition through trademarks is a key component of any AI rollout.

Trade Secret

Trade secrets are a way to protect a company’s intellectual property by keeping it secret. They are an integral part of intellectual property for many companies for know-how and expertise that cannot be easily taken, reverse engineered, or replicated. Further, while trade secrets appear to be an inexpensive way to protect intellectual property, precautions must be taken by the companies to preserve and protect their trade secrets.

Stakeholders in AI should be aware of technologies that are amenable to trade secret protection. AI that operates on proprietary servers, while being used by third parties, can be protected by trade secrets. In addition, trade secrets may be used in conjunction with patents to protect AI. For example, datasets used for training and finetuning AI, as well as the weights applied to data may be held as trade secrets, while the AI models themselves and their structure can be patented. In some instances, the value of AI may be in its output that is not easily replicated. In this case, the AI model that generated AI may be protected with trade secrets.

If companies choose trade secrets as a vehicle to protect their intellectual property, companies must implement and enforce polices and non-disclosure agreements directed to keeping their technology a trade secret. Additionally, companies must provide extensive training to employees about these policies. After all, like a genie in a bottle, once released, trade secrets become difficult, if not impossible to conceal.


Privacy continues to be an important issue for AI, especially relating to the acquisition and use of personal data. Whether it is Internet search data, medical records, or anything number of other sensitive data sources, companies see data as having immense source of value. The more distinct data the company has or can obtain to train its AI, the better AI would be to generating a particular output. At the same time, regulatory environment is uncertain. While forms of privacy regulation and legislation are constantly being proposed at both the state and federal level, privacy regulation largely remains a creature of state, not federal law. This means that enforcement is state specific and affects companies differently in different states. A complaint has recently been filed with causes of action under state privacy law.7 One notable exception is the Health Insurance Portability and Accountability Act (‘HIPAA’) may restrict the use of personal medical records on a federal level.8

Stakeholders in AI systems should be aware that any use of personal data in the training of AI models requires understanding both the current legal environment as well as possible future directions for law and regulation. As states continue to develop their own privacy laws and regulations, companies may have to consider the law of multiple jurisdictions when evaluating data gathering and use. At the same time, companies can also use contract law and have their users agree to provide their data in return for anonymity.


AI may be protected using a variety of intellectual property legal regimes. Deciding on the best way to protect AI innovation requires a fine-grained analysis of each technology and the company’s business. The industry, the type and source of data, business use, and the selection of relevant jurisdictions are all factors that should be considered in crafting a strong company policy to protect AI innovations. After all, it is much more effective to know the legal challenges and risk upfront before a company’s business and revenue are impacted.

This article was co-authored by Sree Gadde, BlueTree Capital Group.

1 17 U.S.C. § 107.
2 See, e.g., Getty Images (US), Inc. v Stability AI, Inc., D. Del., No. 1:99-mc-09999, filed 2/3/23; Tremblay v. OpenAI Inc., N.D. Cal., No. 3:23-cv-03223, filed 6/28/23
3 Available at, starting at 00:19:30.
4 See 35 U.S.C. §§ 271-73, 281 (describing what constitute infringement and the availability of injunctive relief).
5 United States Patent & Trademark Office, Subject Matter Eligibility Examples, pp. 8–9, (2019)
6 See Article 52 of the European Patent Convention.
7 See PM v. OpenAI LP, N.D. Cal., No. 3:23-cv-03199, filed 6/28/23 (listing causes of action under California and Illinois privacy law, among others).
8 Health Insurance Portability and Accountability Act. Pub. L. No. 104-191, § 264, 110 Stat.1936.