USA – MASSACHUSETTS Trends and Developments Contributed by: Paul A. Hughes, Evan S. Kipperman, Daniela Badiola Spanos, Katherine “Katie” Rubino, Tamia Simonis and Mark W. Heaphy, Wiggin and Dana LLP
pre-existing IP that must be excluded from any cus - tomer IP assignment. Customers, however, worry that deliverables are so intertwined with the AI tools that these carve-outs effectively strip them of meaningful ownership. Many providers depend on third-party AI platforms whose terms of service may impose conditions on ownership or use of outputs. Customers increasingly require providers to disclose which platforms they use and to warrant that applicable third-party terms do not conflict with the customer’s IP rights. In all cases, the rapid adoption of AI tools in sourcing arrangements has fundamentally complicated nego - tiations around the ownership of IP. Parties must now address new questions about authorship, protectabil - ity, training-data rights, patentability, third-party plat - form terms, and regulatory compliance. In this new environment, both customers and providers benefit from creating detailed, AI-specific IP ownership pro - visions rather than relying on legacy frameworks that never contemplated AI-generated works. Service level agreements Historically, service level agreements (SLAs) in sourc - ing contracts have been built around human-driven performance benchmarks, response times, resolution rates, processing volumes, and error rates calibrated to the capabilities of human teams. However, the rise of AI has introduced significant new considera - tions and challenges for both customers and provid - ers. When a provider deploys AI-driven automation (eg, robotic process automation, AI-based customer service, or intelligent document processing), the par - ties must negotiate SLAs that address the unique characteristics of AI-delivered services. Negotiating these metrics is more complex than traditional SLAs because AI performance may make bright-line pass/ fail thresholds more difficult to set. Where AI tools are used to deliver or monitor services, questions arise about liability for AI-driven failures, the adequacy of service credits as a remedy, and the circumstances under which AI-related performance failures should excuse or mitigate a provider’s responsibility for such failures.
Service providers are often hesitant to offer service levels for AI tools, citing difficulties around accuracy rates, error tolerances, bias monitoring, explainability requirements, and the inability to control inaccurate or harmful outputs. When an AI system fails or produces an incorrect output, determining fault is far more com - plex than with human error. AI tool performance is heavily dependent on the quality, completeness and timeliness of input data. Providers frequently seek to carve out SLA liability where poor performance is attributable to data supplied or controlled by a cus - tomer. Customers increasingly demand that SLA tar - gets reflect AI-enhanced capabilities, faster response times, near-zero error rates, and continuous availabil - ity, since AI tools can operate continuously without fatigue. Accordingly, the integration of AI into sourc - ing contracts has led to a shift in negotiating SLA provisions, requiring tailored approaches to address the unique characteristics and risks of AI technology while balancing the potential benefits of automation and innovation. Performance warranties Traditionally, sourcing performance warranties have focused on human-delivered services and throughput tied to defined processes. As providers integrate AI tools to deliver or augment those services, the par - ties must reconsider how to define satisfactory per - formance. When an AI system produces inaccurate or substandard output, customers increasingly seek warranties that place responsibility on the provider regardless of whether the issue stems from the model, training data or human oversight. Providers, by con - trast, often aim to limit liability for outcomes that are inherently probabilistic. Customers may also demand transparency warranties requiring the provider to explain how an AI system generated a particular result, while providers resist such obligations, citing trade-secret concerns or tech - nical limitations. The widespread use of third-party AI models further complicates negotiations: customers want the pro - vider to stand behind all outputs, whereas providers often seek carve-outs for failures attributable to AI platforms.
1137 CHAMBERS.COM
Powered by FlippingBook