Considerations when contracting about AI-sytems

With the recent approval of the AI Act by the European Parliament in mid-March, it is crucial to start thinking about the impact of AI systems (and the AI Act) on your organisation and to implement contractual safeguards when entering into agreements concerning AI systems. The complexity of artificial intelligence systems underscores the need for careful consideration and precise drafting of contractual clauses to ensure clarity, fairness, and legal compliance. This article explores the essential considerations for contracting about AI systems.

Importance of the AI Act

One could argue that the AI Act is not relevant for non-high-risk AI systems because almost all of the requirements in the AI Act are focused on high-risk AI systems. If my company is dealing with non-high-risk AI systems, there's nothing to worry about, right? This is not true. Firstly, there are still transparency requirements that apply to the non-high-risk AI systems that your company is using. Secondly, regulated or not, AI systems may still pose significant risks to your organisation, through the form of bias and discrimination, black box and explainability challenges, cybersecurity risks, and data protection concerns. It is essential that your contract ensures safeguards against these risks. Lastly, the requirements for high-risk AI systems (Title 3, Chapter 2 AI Act) do reflect the Ethics Guidelines for Trustworthy AI of the High-Level Expert Group on AI. To ensure that the AI system deployed by your organisation is safe and trustworthy, it is recommended to include (some of) these requirements when contracting about AI systems, and to require your contract partners to comply with them. The fact that these requirements stem from the AI Act and are prescribed by the High-Level Expert Group provides you with leverage to negotiate them into the contract regarding non-high-risk AI systems.

AI Systems

The first consideration is whether you are contracting about an AI system or “regular” software. The AI Act provides a broad definition of an AI system, but there are exceptions where a system does not meet the criteria of an AI system. In such cases, a standard IT contract is appropriate. It is important to recognise the differences in development between regular software and AI systems. Be aware that the development of an AI model is a repetitive process of trial and error, and the outcomes from AI models may not always be precise. This makes it difficult for providers to guarantee performance upfront. Also, be mindful that verifying an AI model is quite challenging and the content and performance of AI models heavily rely on the training data. Finally, remember that the value of know-how is especially significant when developing and utilising AI models.

Different obligations for different actors

Before entering into a contract regarding an AI system or a product that incorporates an AI system, it is important to know what type of actor your organisation is in order to determine what strategy should be followed when negotiating the contract. The AI Act defines four main actors: providers, deployers, distributors, and importers. Each category may have different obligations and responsibilities under the AI Act, emphasising the need for clarity in defining roles and obligations in the contract. Providers are subject to the most obligations, but be aware that in certain situations, deployers, distributors, and importers can be requalified as providers regardless of the contractual qualification. If your company is a provider of non-high-risk AI systems, you may prefer not to be bound by too many contractual obligations, whereas if your company is a deployer, it may be preferable to have these obligations included in the contract.

Complex value chain

It is important to note that the value chain in AI systems is highly complex. In a typical software contract, there is a developer, occasionally an integrator, and a customer. However, with AI systems, it is not just a single provider (developer); there are multiple providers at various levels involved in the creation of AI systems. For instance, during the development of an AI system, providers of tools, services, components, and processes might be involved, alongside the provider of a General Purpose AI (GPAI) model, and the downstream provider. Different contracts will be in place between these providers. Once the AI system is ready, it enters the market. At this stage, various vendors may once again be involved such as the provider itself, importers, distributors, or product manufacturers. Again, another contract will be in place between the developer and the vendor. Subsequently, the vendor will establish a contract with the deployers, who in turn will enter into an agreement with the end user.

Contracting scenarios

Finally, be aware of the different contracting scenarios. Whether you are dealing with a licensing agreement, an integration contract, or a build contract, each scenario presents distinct negotiation margins and risks.

Wouter Torfs