Who Is Liable for Incorrect AI Outputs?

Artificial intelligence assists businesses with tasks such as drafting texts, analysing facts, performing calculations, preparing documents and developing software. Incorporating incorrect outputs into work delivered to customers can cause substantial damage. Liability depends in particular on the contractual relationship, the scope of the obligations and the duties of care applicable to the use of AI.

Contractual Responsibility When Using AI

AI has no legal personality of its own and therefore cannot independently incur liability. In the relationship with the customer, it is generally regarded as a technical tool used by the business performing the contract. The business remains responsible for fulfilling its contractual obligations. Simply attributing an error to the software used is not sufficient to establish an absence of liability.

The first question is what performance is owed. Under an agency contract, the primary obligation is the diligent and faithful performance of the business entrusted to the agent (Art. 398 para. 2 of the Swiss Code of Obligations, CO). An adverse outcome does not, in itself, give rise to liability. The decisive question is whether the business exercised the care required in the circumstances. Adopting an AI output without verification where verification is required may constitute a breach of that duty.

Under a contract for work and services, by contrast, the obligation is to achieve a result that conforms to the contract. Where, for example, bespoke software is commissioned under such a contract, it must possess the required characteristics, regardless of how the code was produced. In the event of defects, the customer may be entitled to rectification, a reduction in price or refusal of acceptance, subject to the applicable requirements (Art. 368 CO). These remedies generally do not require fault. Contractual provisions and requirements concerning inspection and notification of defects must also be observed. Any claim for damages must be considered separately, as it is subject to its own requirements.

The specific obligations always form the starting point. The use of AI does not create a general guarantee of success or quality, nor does it release the business from its existing obligations.

Requirements for Liability in Damages

Under the general rules on contractual liability, a claim for damages requires a breach of contract, damage, and a natural and adequate causal link between the two (Art. 97 CO). Fault is presumed; the business may, in principle, avoid liability by proving that it was not at fault. Specific liability and warranty provisions remain applicable.

Due Care in Selection and Review

The requirements governing the use of AI depend on the task, the foreseeable risks of error and the potential consequences for the customer. A professional recommendation with significant implications may require detailed verification, whereas different checks may suffice where AI is used merely as a drafting aid.

Businesses should therefore assess whether an application is suitable for its intended purpose and how its outputs can be reliably verified. Depending on the risk, this may require consulting original sources, independently checking calculations or involving suitably qualified professionals. Properly documented checks can help establish, in the event of a dispute, that the required care was exercised. They do not, however, replace either the verification actually required or the performance owed under the contract.

Attribution of Liability and Recourse

Subject to the statutory requirements, the business must also answer for the conduct of persons, particularly employees, whom it engages to fulfil its contractual obligations (Art. 101 CO). This must be distinguished from the employee’s liability towards the employer under Art. 321e CO, which takes account, in particular, of occupational risk and technical knowledge.

Recourse against the AI provider likewise requires a separate legal basis. The provider’s contractual obligations and any limitations of liability are decisive. The business cannot therefore simply pass on its own liability towards the customer to the provider.

Conclusion

The use of AI changes how contractual obligations are performed, but generally not the obligations owed to the customer. Whether, and to what extent, a business is liable depends on the performance owed and the requirements of the particular claim. Clear contractual specifications and quality controls proportionate to the risks are therefore essential. Potential recourse against employees or providers does not relieve the business of its own responsibility towards the customer.