Contract manufacturing organizations (CMOs) engage machine learning in their processes. Such machine learning systems increase their capacity based on data that are often exclusive property of the commissioner according to the contract. Moreover, CMOs are required not to use for competitors the knowledge acquired through the work for the commissioner. Nonetheless, machine learning, trained by the work for a commissioner, are afterwards employed for different commissioners, often including competitors. The paper evaluates whether CMOs can rely on the concept of know-how to justify their trained machine learning, and it concludes that such solution could be acceptable from a policy perspective if specific conditions apply.