“Keeping Proprietary Data Out of AI Training Models”

Jake Vollebregt was interviewed by CIO.inc for its Q&A article, “Keeping Proprietary Data Out of AI Training Models.” The article features Dykema Senior Counsel Jake Vollebregt discussing the growing legal and business risks associated with vendors using enterprise data to train artificial intelligence models. Vollebregt explains that awareness of AI training data risks has increased as businesses have incorporated generative AI into their operations, but organizations still face challenges when AI features are embedded within existing vendor services. He emphasizes that protecting proprietary information requires collaboration among legal, procurement, information security, privacy, and business teams rather than relying on a single department.

Vollebregt explains that the primary concern for businesses is intellectual property leakage, including the potential loss of trade secrets, proprietary processes, code, business strategies, and customer insights when data is incorporated into AI training sets. He compares the development of AI data protections to the evolution of cloud and SaaS contracting practices, noting that organizations are beginning to establish more standardized approaches. He recommends that companies require vendors to put explicit “no training on our data” commitments in their contracts or signed addenda, rather than relying on dashboard settings or marketing promises, and stresses the importance of technical controls, auditing, and verification to ensure those protections are actually enforced.