In this episode of Corruption, Crime and Compliance, Michael Volkov explains why standard software procurement templates fail to protect organizations in AI vendor relationships, and what to do about it. He walks through the structural differences that make AI vendors riskier than traditional software vendors, multi-layered data flows through underlying foundation models, frequent model swaps, and vendors’ commercial incentive to train on customer data, and identifies the specific gaps legacy contracts leave open: silent or vague data training rights, indemnification that doesn’t reach model outputs or training data provenance, missing audit rights, and liability caps that quietly undercut existing protections. The episode closes with a concrete negotiating playbook, including explicit training restrictions, coordinated indemnification and liability provisions, audit rights, subprocessor disclosure, regulatory compliance representations, and guaranteed exit and data deletion rights, along with practical guidance for organizations facing dominant vendors unwilling to negotiate.