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
AI adoption has also created new categories of per - formance warranties absent from traditional sourcing. Customers increasingly require assurances that AI tools will not produce biased, discriminatory or other - wise unlawful outputs, particularly in HR, financial ser - vices and healthcare. Customers also seek enhanced warranties around system security and protection of input data, due to security risks introduced by AI sys - tems, and they want assurances that AI-generated outputs do not infringe third-party IP rights, a con - tentious issue amidst ongoing generative-AI litigation. Negotiating performance warranties has become much more complicated and the appropriate warranty terms for any sourcing arrangement will necessarily depend on the services involved, the AI tools used, the regulatory environment and each party’s risk tol - erance. Indemnification provisions The rise of AI tools in sourcing arrangements has significantly reshaped how indemnity provisions are negotiated. Customers increasingly seek broader infringement indemnities that expressly cover claims arising from the provider’s use of AI tools, while pro - viders push to limit liability for risks they view as inher - ent to AI systems and outside their control. AI tools require access to substantial datasets, often including customer confidential information or per - sonal data. This has intensified negotiations over data breach and privacy indemnities. Customers worry that inputting their data into third-party or cloud-based AI platforms could lead to unauthorised disclosure, data leakage, loss of trade-secret protections or regula - tory violations. Customers extend indemnity clauses to cover losses stemming from AI-driven data pro - cessing, but providers seek to limit these obligations to instances of their own negligence or breach rather than accepting strict liability for failures of underlying AI platforms. Where providers use AI to support professional or advisory services, such as legal research, financial analysis or technical recommendations, customers increasingly request indemnities for losses caused by inaccurate or “hallucinated” AI outputs. This differs from traditional sourcing indemnities, which contem -
plated human error rather than systemic algorithmic errors. Negotiations often focus on the provider’s duty to maintain human oversight and quality controls, with indemnity scope turning on whether the provider can demonstrate adequate review processes for AI-assist - ed work. Many providers rely on third-party AI platforms rather than proprietary models and thus argue for a layered risk structure. Customers typically argue that provid - ers should bear full responsibility for the tools they choose, including third-party AI, and should offer “flow-down” indemnities regardless of whether they can recover from the AI vendor. Providers contend that certain risks, such as fundamental flaws in widely used AI models, should be treated as force majeure or excluded from indemnity obligations altogether. In sum, the integration of AI tools into sourcing arrange - ments has expanded both the scope and complexity of indemnity negotiations, driving the development of new indemnity categories and intensifying disputes over allocation of third-party risk. Liability limits The growing use of AI tools by service providers has reshaped negotiations around limitation-of-liability clauses. Historically, these provisions focused on service failures, data breaches and IP infringement. However, AI introduces a broader set of risks. When providers rely on AI to deliver services, such as analyt - ics, automated decision-making or AI-assisted cod - ing, the possibility of inaccurate, biased or irrational outputs creates new liability exposure. Customers often argue that these risks should fall outside stand - ard caps or be treated as carve-outs. If AI produces discriminatory outcomes in areas such as HR out - sourcing or claims processing, customers may face regulatory scrutiny and reputational harm, prompting demands for uncapped liability or higher super-caps. AI-driven training on customer data adds further con - cerns around data protection, confidentiality and IP ownership, all of which influence liability discussions. Customers often contend that the potential scale of harm – regulatory fines, reputational damage and third-party claims – can far exceed contract value, increasing pressure on providers to accept higher
1138 CHAMBERS.COM
Powered by FlippingBook