As artificial intelligence ("AI") continues to grow and pervade our consciousness, more technology, health care, financial services, and manufacturing businesses are deciding whether a capital investment in or acquisition of AI technology owned by another business is the best path forward. Below are certain legal considerations in contemplating an acquisition of certain AI owned by another company via a stock or an asset purchase.
Conduct AI-Specific Due Diligence
While the concept of AI has been prevalent for decades, M&A attorneys may not be familiar with the underlying components or terminology that creates the AI. A thorough understanding of these aspects is imperative, and the buyer's deal counsel should include subject matter experts to conduct due diligence in order to identify and understand the AI components. This process will also assist the buyer in determining the seller's most valuable assets and allow for the buyer's counsel to appropriately define the various aspects of the AI components for risk assessment purposes.
Ultimately, the key to AI is the input data provided to the AI system that, in turn, continuously trains and teaches the AI system based on the data provided. Just as important are the ultimate rights to this underlying data. Therefore, the buyer must understand whether there are any inherent biases in the data provided to the AI system. Any biases could ultimately create an AI output that itself is biased, which may devalue a buyer's use of such outputs in the future.
Be Mindful of Definitions
Understanding the building blocks of the seller's AI may include the following concepts and definitions:
- An "algorithm" is a set of rules a machine can follow.
- "Machine learning" is a subset of AI that enables computers to improve their performance based on the data provided and without any specific programming goal.
- A "large language model" or "LLM" is typically the data used to train the AI system and usually includes multiple algorithms.
- "Generative AI" is AI that appears to generate new content (e.g., text, images, audio). Such content is derived from the large language model training the AI.
Ownership of the Components
Determining whether seller solely owns the AI components, or whether a portion of the components are licensed from one or more third parties, is often key. It is likely that seller has licensed an algorithm or developmental product in order for the AI to be created. The buyer must understand all of the components that, in the aggregate, create the AI output. Sometimes, the seller does not have proper rights to this underlying property.
Further, the buyer must determine that all current and former employees, as well as any service providers or independent contractors, involved in the creation or development of the AI components have assigned all of their individual ownership rights in the components to the seller. The buyer's counsel should confirm whether all employees or independent contractors have fully executed and enforceable Invention and Assignment Agreements or whether an employment agreement or independent contractor agreement includes a provision that requires the individual to assign all ownership rights in the AI to the seller.
Representations and Warranties of Seller
Seller's representations and warranties in the definitive agreement should include a corresponding schedule requiring the seller to list all AI components and outputs and whether such items are owned directly by seller or licensed by a third party. If any technology used by the seller is hosted by a third-party provider (i.e., cloud service provider or application service provider), the buyer should include a representation that the seller is in compliance with all terms and conditions for the licenses it has obtained in order to use such technology. Buyer's counsel should also carefully review these license agreements to confirm the seller has complied with all terms and conditions and determine any additional obligations the buyer may inherit, such as the scope of indemnification provisions.
The representation should also include the seller's:
- possession and the right to license or use the AI components;
- non-infringement of any intellectual property rights based on the current use of the components;
- confirmation that there are no material errors, defects or failures in the components or in the use of any personally identifiable information used to train the AI; and
- confirmation that the components conform to industry standards and practices.
The buyer should contemplate including a representation regarding open source software. Open source software is software distributed in source code and made "open" to the public to use, modify and distribute the source code for any purpose. Open source software would be considered an "open" generative AI system. Typically, a corresponding license has been provided to the seller to use the open source software. This representation will require the seller to confirm that the open source software does not require the distribution of any proprietary software in order to use the open source software, and that the seller has exclusive control over the source code as it relates to its proprietary software. Particularly, the buyer should conduct due diligence to confirm that none of the seller's confidential information has been included in the open generative AI system, as such information would continue to be available and allow competitors to access confidential information.
In addition to reviewing the seller's cybersecurity insurance policy, the seller should also include a cybersecurity representation that requires the seller to represent to material compliance with:
- data security and data privacy laws, including any contracts related to the collection, use, storage and processing of personally identifiable information ("PII"); and
- all security measures and policies related to PII, including contracts that require certain security requirements regarding PII.
Regarding indemnification, the buyer should assert that any representation related to AI and intellectual property is considered a "fundamental" representation in order for these representations to survive for a longer duration after the closing date and possibly be excluded from any indemnification cap. Buyer and seller's counsel will negotiate the survival period of the non-fundamental representations, which typically last 12 to 18 months after the closing date.
AI may be subject to specific state, federal and international data privacy laws, which are outside the scope of this article.
Conclusion
As AI continues to be a focal point, expect to see more acquisitions of these companies. Comprehensive due diligence and careful lawyering can help mitigate the risks of any acquisition.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
