We propose a measure for genuine multipartite correlations suited for the study of dynamics in open quantum systems. This measure is contextual in the sense that it depends on how information is read from the environment. It is used to study an interacting collective system of atoms undergoing phase transitions as external parameters are varied. We show that the steady state of the system can have a significant degree of genuine multipartite quantum and classical correlations, and that the proposed measure can serve as a witness of critical behavior in quantum systems.PACS numbers: 03.65. Ud,03.67.Mn,05.70.Fh Introduction.-There is a lot of interest in the characterization and quantification of correlations between quantum systems. The fact that quantum systems can be correlated in ways that classical systems can not is generally understood to be essential to emerging quantum technologies [1], and thus understanding the nature of such correlations is one of the most active areas of research in quantum information theory. Recently, there has been much effort made towards quantifying correlations-quantum or classical-for multipartite (as opposed to bipartite) systems. In particular, there exists a question of whether or not a system contains genuine multipartite correlations, i.e., are there correlations that cannot be accounted for by any subsystem alone. A general framework, where a measure for a type of correlation was defined as the "distance", as measured by relative entropy, from one state with a defining property to the closest state without said property, was recently proposed by Modi et al. in [2]. This framework permits a unified view of the measures of relative entropy of entanglement [3], quantum discord [4], classical correlations [5], quantum mutual information and several new quantities. A powerful feature of this framework is that the suggested measures are well defined, and relations between them also hold for multipartite systems, thereby allowing for natural extensions of the above quantities beyond the bipartite case. A natural follow up to the work in [2] was done by Giorgi et al. in [6] where n-partite quantum and classical correlations were further divided into genuine and non-genuine parts. This is not the only effort in this direction however, and it is not known how to correctly measure genuine multipartite correlations. Investigations of the validity of the various measures involving the dynamical evolution of correlations in experimentally relevant situations therefore seem beneficial. This is the purpose of the work presented here.