By Jill Hubbard Bowman
Introduction
A license is merely permission to act. Usually, the owner of an intellectual property right (like a copyright) to the technology grants a license (permission) to another company to do something with the technology it can’t legally do without permission. If a copyright exists to a work of authorship, it gives the owner the right to restrict others from reproducing, making a derivative work, distributing, publicly performing, or displaying the copyrighted work. An owner can license one or all of the rights.
Other intellectual property rights protect different subject matter. These rights also differ in scope of protection. A patent grants broader rights to inventions. It gives the owner the right to restrict others from making, using, selling, offering to sell, or importing its invention. A trade secret right protects information. Its scope is broader and narrower. It gives the owner the right to restrict others from using or disclosing the trade secret.
In software agreements, especially open-source licenses, developers have traditionally relied on ownership of copyrights to the software for the license grant and to control what a party can do with their software but . . .
- What if the AI technology being licensed isn’t protected by copyrights?
- What if someone else owns the copyrights (or other IP rights)?
- What if modified materials can’t be copyrightable derivative works?
AI technology doesn’t fit copyright law or traditional licensing frameworks.
![](https://ailawmazemap.com/wp-content/uploads/2024/06/Firefly_23eb8d0b-3974-4fa8-afca-20ed93aa9e35-1024x796.jpeg)
The major paradigm shift caused by AI technology has radical implications for using copyrights (against the world) and in licenses (against the other party) to control access, use and monetization of AI technology. Companies on both sides of a deal need to deeply consider whether either party, or anyone else, has copyrights to the specific types of AI technology being licensed and modified in a deal (around the world). [Then, they need to figure out what other types of IP rights might exist, which is beyond the scope of this article.]
The special characteristics of AI technology, including its ability to morph autonomously as it continues to learn and create output, creates more licensing challenges for understanding what intellectual property rights arise, if any, and how those rights and potential ownership may change as the technology shifts across country borders.
In the U.S., copyrights can’t be relied on in AI technology licenses.
Traditional technology license agreements are structured on several big assumptions. First, the biggest underlying assumption is that there is 1) technology, and 2) that technology has IP rights that restrict what the world can do with it. Definitions, license grants, and other terms are based on this assumption. Traditional software licenses rely on copyrights. There is a foundational assumption that the licensed software is protected by copyrights.
Most AI technology, however, is not like regular software. What is most valuable is often a compilation of facts or works by third parties, algorithms used to create and process the AI technology, algorithms embodied in AI technology, technical improvements for efficiency or accuracy, or material autonomously generated by a computer-technology that may have thin insufficient protection or more often, no copyright protection at all under U.S. copyright law.
The first step to dealing with copyright issues in licensing AI technology is recognizing copyrights may be missing.
It’s important to determine:
- Is the AI technology protected by copyrights?
- What rights, if any, may arise to modifications?
- Who owns the copyrights if they exist?
You might be surprised by the answers.
(See the previous U.S. Copyright Law posts on AI Law Maze Map for more information about AI technology and copyrightability and derivative works.)
Modified AI technology may not be a copyrightable “derivative work.”
The second big assumption made in traditional technology agreements is that modification of the licensed technology, like software, gives rise to a copyrightable derivative work, which has a subsequent copyright that arises automatically by operation of law upon fixation and can be owned and used for downstream protection of the modified technology.
In AI technology agreements, the ambiguous term “derivative work” may be taken out of the statutory copyright context and loosely used to mean any modified AI technology, regardless of whether the pre-existing technology or subsequent technology is copyrightable. The use of the term “derivative work” implies that the modifications are new copyrightable derivative works. The agreement may even say that the licensee “owns” derivative works created with the licensed AI technology. But nobody “owns” a copyright to uncopyrightable AI technology or subsequently uncopyrightable modifications. For some important AI technology, these false assumptions may result in the failure to take appropriate measures in contracts to achieve the business’ objectives of controlling access, use, or monetization the AI technology. The license grant may be to rights that don’t exist. Restrictions may be part of the license to nothing instead of clear and in a separate term. Further,modifications of even a copyrightable software work may not be a copyrightable derivative work because the modifications are purely technical or created by a computer without human control.
Similarly, the term “derivative data” may be used to refer to uncopyrightable modifications to uncopyrightable database data as a spin on the statutory derivative work copyright concept. This term may be very broad, vague, and ambiguous. It may be used as a tactic to gut contractual restrictions on the use of licensed data and shift control of the data to the licensor.
A lack of a copyrightable derivative work may create a kink in downstream licensing schemes. Developers and deal makers need to understand that without IP rights and in the absence of contractual or other legal restrictions, anyone can use the technology. They also need to understand typical contractual restrictions are only between the two deal parties-not the world. Open-source licenses, however, are a special case and may be a contract with someone who has notice of the contract and uses the software. [i] But in the absence of a copyright, the threat of loss of the copyright license, which is the foundation of the open-source software ecosystem, is undermined.
Without copyrights, it's possible to “license” AI technology and its modifications based on a contractual agreement or other IP rights if they exist. It’s possible to use a functional license grant that does not refer to IP rights. The contract can just clearly say what the parties can and cannot do with the technology. It’s also possible to define a type of contractual “ownership” if legal property rights are missing. Further, a licensor can require a licensee to impose contractual terms on downstream users.
In summary, when licensing AI technology, traditional assumptions about copyrightability and the existence of copyrightable derivative works should be questioned. Don’t just try to shoe-horn new AI technology into old agreements. More analysis is needed if a company wants to control access, use, and monetization of AI technology. Traditional agreements and term sheet templates should be modified to reflect the parties’ agreement in the absence of copyrights.
[i]Rothman, Marc, “Breach of Contract Claims Allowed for Open Source License Violation,” https://www.dglaw.com/breach-of-contract-claims-allowed-for-alleged-open-source-license-violations/