Introduction
It is widely accepted that information sharing may significantly benefit supply chains , especially in reducing the bullwhip effect , decreasing supply chain costs, and increasing the efficiency of collaborative product developments . For this reason, it is common for a company in a supply chain to share a large amount of information with its partners.
However, information sharing in supply chains is also a doubleedged sword: it may have an adverse effect, namely, information leakage . In general, information leakage means confidential information is unintentionally revealed to unauthorized parties. In supply chains, information leakage is a serious threat due to real incentives, that is, companies have strong motivations and more than enough capabilities to collect, analyze, acquire, and utilize information from others to gain a competitive edge. 1 Information leakage in supply chains can take two different forms. First, confidential information may be mistakenly shared, resulting in the so-called direct information leakage. To avoid direct information leakage, companies need a precise answer to the question: What information is confidential? However, providing such an answer is usually more challenging than it looks, as we shall show shortly. Second, confidential information may also be unintentionally leaked in the form of inferences. An inference happens when confidential information can be inferred from other, seemingly nonconfidential, shared information. This is possible due to the inherent engineering relationships between different pieces of information. To prevent damaging information leakage caused by inferences, companies need to answer the question: What inferences are possible, and what is the risk of information leakage caused by such inferences?
Unfortunately, answers to the questions above provided by most existing technical solutions are not fully satisfactory (the detailed review of existing solutions is given in Section 2). First, those solutions typically assume the classification of confidential information is already given, or can be trivially obtained. However, this is not always true. Different pieces of information in a complex engineering design, such as different parameters appearing in the same engineering formula, usually have an entangled relationship. Such a relationship may blur the boundary between confidential and non-confidential information. Second, to our best knowledge, the possibility of potential information leakage caused by inferences is simply ignored in most existing technical solutions (a few exceptions will be discussed in Section 2).