Contracting for AI applications
Things to know
- Given the nature of artificial intelligence, particularly large language models, most companies seeking to derive value from AI will procure — not build — AI systems. Effective contracting that takes into account the particularities of AI technologies is critical.
- Contracting frameworks for information technology generally are used as the basis for acquisition or sale of AI products and related services, but many unique attributes of AI products and services require that particular attention be applied to AI-specific issues.
Things to do
- Delineate the components of an AI system, which may include foundational models, fine-tuned models, algorithms producing models, data, software applications and interfaces, and agents.
- Allocate rights to the various components of the AI system; this often requires going beyond a simple allocation based on products, existing or background intellectual property and ownership of new works, to more complex discussions about grants of license rights, restrictions on use and other factors on the various components comprising the AI system.
- Account for compliance issues by allocating related responsibilities and including mechanisms to deal with changes arising from quickly developing regulatory requirements and standards.
- Address issues relating to data that are inherent in AI systems, such as data used to train models, use and disposition of customer data, query data, and outputs.
- Incorporate necessary elements of responsible AI into the contractual arrangement, including, as applicable, model accuracy testing and improvement, transparency, and bias testing.
- Include commitments on reliability and availability commensurate with the importance of the business function being enabled by the AI product or service.
- Allocate risk by way of warranties, indemnities and limitations of liability through a risk-based approach to creation, use and management of the AI system. Carefully account for the risks associated with the specific AI system distinguishing between the system itself and the outputs.
Useful resources
- “Navigating the legal landscape of AI commercial contracting,” Osler, September 17, 2024
