AI Clauses in Client and Subcontractor Contracts: What to Include to Protect Rights, Data and Liability

Where Disputes Most Often Arise

In AI implementations, disputes usually concern not the model itself, but who is responsible for the result, who may use the outputs, and whether client data may be used for further training. In practice, these risks need to be described in the contract already at the scope-of-work stage, because according to Sawaryn I Partnerzy, a lack of precision in these areas leads to disputes over rights to outputs, improved model weights and control over the quality of results.

According to Creativa Legal and Sawaryn I Partnerzy, it is also worth separating the roles in a project: who is the provider of the AI system, who is its operator, and who bears the consequences of the tool's operation in a specific business process. This matters especially when the client expects not only an implementation, but also the use of the tool in a decision-making or production process.

Clauses for the Client Contract

The client contract should include at least three elements: the scope of permitted AI use, the rules for accepting outputs and the division of rights to the result. Sawaryn I Partnerzy point out that it is also useful to define "AI outputs" as output data covered by a licence or trade secret protection, rather than automatically by copyright.

The practical wording of these provisions should answer the following questions:

1. Is AI merely an auxiliary tool, or may it create elements delivered to the client?
2. Who holds the economic rights to the output: the client, the contractor, or both parties on separate terms?
3. Does the contractor verify the output for plagiarism, errors and unauthorised modifications of third-party works?
4. Does the client receive the right to reuse the output in its own business?

How to Secure Subcontractors

With subcontractors, it is essential to enforce full transparency regarding AI use. According to attorney Bogna, a marketing agency should require freelancers, in their contracts, to disclose in writing whether and to what extent they used AI tools when creating material. This prevents a situation in which the commissioning party takes responsibility for content generated without knowledge of how it was created.

Sawaryn I Partnerzy, for their part, recommend that implementation contracts expressly prohibit the use of AI models without the prior consent of the commissioning party or, at a minimum, assign the contractor full responsibility for the consequences of such use. In the relationship with a subcontractor, it is also worth adding a prohibition on training models on client data, an obligation of confidentiality, and an obligation to disclose which tools were used.

What to Add in Projects Covered by the AI Act

If the solution may fall under the AI Act, the contract should determine which party is responsible for the classification of the system, technical documentation and labelling of AI-generated content. Inink emphasises that the absence of a clear declaration of classification means that the deploying party may itself bear the consequences of an error.

In practice, it is also worth adding an obligation to cooperate in monitoring the system's operation, reporting incidents and updating the contract provisions when regulations change. This is not legal decoration, but a mechanism that keeps the contract aligned with a solution whose risk profile may change after the contract is signed.

Lub System helps B2B companies implement AI, automation and IT solutions end-to-end - from strategy to deployment. See our services or get in touch to discuss your case.