The rise of AI assets and what it means for intellectual property due diligence in M&A transactions

Now more than ever, artificial intelligence (AI) is on the rise. Following the recent release of several highly advanced AI models, we are seeing immense growth not only in AI-based applications, but also in content developed by these applications, so-called "AI assets". These AI assets will play – and are already playing – an increasingly important role in transactions, especially with regard to questions like who is the owner of the AI asset and who is liable for the creation and the use of the AI asset. However, most of the currently used intellectual property (IP) due diligence request lists do not significantly mirror the significance of such AI assets of the target company. Therefore, the following article provides insights on specific AI-issues relevant for IP due diligence processes.

Why the Market for AI is growing rapidly

AI is increasingly affecting the whole business world. Current forecasts expect the market for AI to grow at a rate of 37.3 % annually in the next seven years. By 2030, the market for AI is expected to reach a revenue of USD 1,345.2 billion as compared to USD 86.9 billion in 2022. The recently released GitHub Octoverse report supports these numbers. According to the report, generative AI projects using pre-trained AI models have explosively grown in 2023 and we may now consider use of AI software as mainstream among software developers.

This is due to the increasing availability of AI. Many Tech Companies like OpenAI, Google, Microsoft and Amazon are rolling out AI platforms that enable users to build custom AI systems. Just recently, in early 2024, OpenAI launched their custom GPT store. Microsoft, Google and Amazon are following closely behind with building-platforms like Copilot Studio, the Gemini-based Android AICore and Amazon Q. These platforms promise customers the ability to easily build custom AI applications for specific purposes with near endless possibilities. In addition, many companies are developing their own AI based tools. The consequence of this is that AI becomes broadly available, and many companies will have the desire and the need to incorporate AI into their business models in order to stay competitive or to improve their competitiveness in their respective markets. Therefore, the amount of M&A transaction which involve AI assets will increase significantly over the next decade and will– in the near future – – not be the exception, but the rule in all M&A transaction.

How this affects IP-Due Diligence in M&A Transactions

The growing market for AI assets will significantly affect the IP due diligence process in M&A transactions with the need to amend existing IP due diligence requests lists and the introduction of AI specific warranty and liability/indemnification clauses in both share or asset purchase agreements.

AI is a complex technology in the first place. How companies will handle questions like “who owns the AI program?", "can the outcome created by AI be classified into a known category of IP rights?", "who owns the outcome created by the AI?", and "who is liable for it?” is even more complex. The legal implications are broadly unclear and may differ from country to country. Therefore, this gets even more complex when we have a transaction which plays in an international setting – which is not the exception but the rule after so many years of globalization.

Against this backdrop, we list in the following the nine most important Dos which should be considered by both deal parties during the IP due diligence process in M&A Transactions when dealing with AI assets:

1. Identify AI assets. AI is more than simple software code. It includes (i) the AI algorithm, (ii) AI training data, (iii) a trained neural network, (iv) its topology, (v) websites, apps and interfaces, as well as (vi) the outcome created by the AI. All these components may be subject to independent IP rights and thereby represent different AI assets. Companies should thus take particular care to identify which AI assets are relevant to the deal in order to be able to identify in a next step which particular IP rights and corresponding legal rules are concerned.
2. Clarify IP Protection of AI assets. The task to clarify if and how the AI asset is protected as an IP right can be difficult since the rules to protect AI assets can vary country by country and while local IP rights protection ends at the borders, the use of the AI will certainly not.

For example, in some countries, it is possible to get patent protection for software, while other countries do not provide patent protection for software as such but only for the method implemented by the software, or for specific parts of a computer-implemented invention. The source code of AI assets as such should be protected under copyright law in most countries whereas the trained neuronal network, its topology, or the core AI algorithm as such are – usually – not eligible for patent or copyright protection.

The same applies for the most valuable part of any AI: the trained AI model (i.e., the object code including the weighted nodes of a neural network) which is usually protected as a trade secret.

3. Examine protection measures for trade secrets, IT infrastructure and cybersecurity. As patent or copyright protection may not be available to key AI assets, the owner of the AI asset can protect these assets as trade secrets, provided that he applies suitable protection measures to keep the trade secret confidential which is required by most national trade secret laws around the world. In order to evaluate the protection status as well as the value of the AI asset, it should be reviewed if and how the AI asset is protected by specific protection measures. In this context but not only for the purpose of sufficient trade secret protection, it should also be evaluated whether the IT infrastructure and the cybersecurity system is sufficient to avoid and prevent e.g. third party attacks from outside the organization.
4. Review ownership and chain of title. The development of AI assets is usually a team effort and many people are included in the creation, design, development, editing, or enhancement of AI assets. Therefore, several firms or individuals may hold ownership rights as creators of the underlying AI technology. In the context of IP due diligence, it is therefore of particular importance to verify who is the current holder of all necessary IP rights in the AI asset and if – especially if the asset has been developed by a third party – the chain of title of the asset from the creator to the current holder can be sufficiently demonstrated.
5. Watch out for Open-Source Software. The use of Open Source Software is common in any software development, so this is a check point for any software due diligence. As pre-trained open-source AI models will become more and more available, software designers increasingly will rely on Open-Source Software (OSS) to build AI applications. Therefore, it is essential for IP due diligence process to identify all relevant license terms of Open-Source Software used in order to verify which specific preconditions might be applicable and whether potential copyleft effects applicable to (parts of) the software could restrict the use of the AI asset.
6. Determine inbound and outbound license agreements. Nowadays the development of IP is often not a “one man”- job and many different companies might be involved in joint developments or might be contributing their own IP to another companies’ project. Therefore, it is essential to check whether IP rights by third parties have been licensed which are necessary to use the AI asset and/or if license arrangements have been made concerning the AI asset which may restrict the future use of the AI asset.
7. Evaluate AI-generated content and its possible IP protection. Not only the AI owned by a company can be of interest in the M&A transaction. Often the AI creates own work products which in the M&A transactions beg the question (a) who is the owner of the work product (e.g., the company creating or running the AI?), and (b) how this AI generated work product can be protected. Especially, the latter question can be very difficult to answer since most national laws are unclear and – at least for the moment – it is not possible to receive a copyright and very likely not a patent for AI generated content, since the AI can be neither an inventor nor an author. The only way to protect AI created content is currently therefore as a trade secret which requires the review of the protection system in place in order to ensure sufficient protection in the first place.
8. Check for infringement of third-party IP rights. The question whether the IP asset is infringing third party IP rights is one of the most essential questions during the due diligence process. This is an essential question for the AI asset as well and ongoing litigations or potential known claims of third parties must be reviewed thoroughly. In addition, the training of AI models requires vast amounts of data which can as such also potentially infringe third party rights, especially trade secrets and copyrights, if no relevant license agreements are in place.
9. Consider implications of regulations. AI moves more and more into the focus of the various regulators. Selling and/or purchasing AI assets might come with obligations addressed by individual laws. In the European Union, for instance, the Digital Markets Act, the Data Act or the AI Act could have implications for both the buyer and the seller before and after the transaction. Cyber security may also play a significant role during the due diligence process, as well as GDPR compliance.

To conclude, it is essential to adjust the due diligence process in a M&A transaction when AI assets are involved (in both asset and share deals). Where the due diligence process reveals material risks with the risk of the use of an AI asset, this can impact the valuation and potentially dissuade a buyer from concluding the deal. Of course, this is not only specific to transactions involving AI assets, but the overall risk assessment changes, since so many legal implications concerning AI are currently unclear. Therefore, at least for the moment, it is essential that the due diligence process is carried out properly and certain risks are mirrored in the respective representation and warranty clauses in the corresponding SPAs/APAs.

This was only a high-level overview of the most important IP issues surrounding AI assets in M&A transactions. Please reach out to the authors of this article if you are interested in further insights or need any support with your IP due diligence process.

Authored by Anna-Katharina Friese-Okoro, Fabian Flüchter and Benedikt Dellen.