We propose a novel approach for statistical risk modeling of network attacks that lets an operator perform risk analysis using a data model and an impact model on top of an attack graph in combination with a statistical model of the attacker community exploitation skill. The data model describes how data flows between nodes in the network -- how it is copied and processed by softwares and hosts -- while the impact model models how exploitation of vulnerabilities affects the data flows with respect to the confidentiality, integrity and availability of the data. In addition, by assigning a loss value to a compromised data set, we can estimate the cost of a successful attack. The statistical model lets us incorporate real-time monitor data from a honeypot in the risk calculation. The exploitation skill distribution is inferred by first classifying each vulnerability into a required exploitation skill-level category, then mapping each skill-level into a distribution over the required exploitation skill, and last applying Bayesian inference over the attack data. The final security risk is thereafter computed by marginalizing over the exploitation skill.