Online Negative Databases

Abstract

Abstract. The benefits of negative detection for obscuring information are explored in the context of Artificial Immune Systems (AIS). AIS based on string matching have the potential for an extra security feature in which the "normal" profile of a system is hidden from its possible hijackers. Even if the model of normal behavior falls into the wrong hands, reconstructing the set of valid or "normal" strings is an N P-hard problem. The data-hiding aspects of negative detection are explored in the context of an application to negative databases. Previous work is reviewed describing possible representations and reversibility properties for privacy-enhancing negative databases. New algorithms are described, which allow on-line creation and updates of negative databases, and future challenges are discussed.

Hiding a Needle in a Haystack Using Negative Databases

“…This speaks to how easy it is to obtain a valid string given a retrieved entry from DB . An intractable amount of strings in DB is not enough to guarantee security unless the number of distinct texts (and codes) is also intractable (it is easy to guide the search away from a few selected strings by including them in N DB (see [10])). The security of the present scheme relies on including an intractable number of strings with distinct texts and codes.…”

“…In contrast to [7], our proposal does not rely on cryptographic primitives for the security of the system, and its construction allows for the stored data to be manipulated via a special algebra (see [13]). Reference [10] discusses several algorithms for updating a negative database once it has been created.…”

Protecting Data Privacy Through Hard-to-Reverse Negative Databases

“…As an example, we focus on the model introduced in [32] and use it as a basis for creating N DBs. This differs from the algorithms described in [20,18,21,17] in two ways: First, it generates an N DB for each string in DB; And second, it creates an inexact representation of U -DB, meaning that some strings in addition to DB will not be matched by N DB.…”

“…In [20,18,21,17] several algorithms were given that either produce N DBs that are provably easy to reverse, i.e., for which there is an efficient method to recover DB, or that have the flexibility to produce hard-to-reverse instances in theory, but have yet to produce them experimentally. It was shown in [20] that reversing a N DB is an N P-hard problem, but this, being a worst case prop-erty, presents the challenge of creating hard instances in practice.…”

“…In our construction, we will use one SAT formula to represent each record in the positive database. The approach is to start with an assignment s (a binary string representing the truth values for the variables in the formula), and then create a formula satisfied by itmuch like the algorithms in [20,18,21,17], except that the resulting formula might also be satisfied by other unknown assignments. Given the assignment s, the algorithm randomly generates clauses with t > 0 literals such that each clause is satisfied with probability proportional to q t for q < 1 (q is an algorithm specific parameter used to bias the distribution of clauses within the formula).…”