Critical infrastructures such as the power grid, up and down stream components of oil and gas production, communication network, transportation networks and so on are now all controlled by devices with CPUs and memory that communicate over both wireline and wireless channels. Quantitative assessment of risk to the controlled infrastructure depends both on models and analysis of the infrastructure under attack, and of the adversary's ability to mount those attacks. To quantitatively assess the risk to the critical infrastructure of cyber-mischief we have to be able to quantitatively assess that component of the risk which depends on the adversary's cyber access to devices which measure and manipulate the physical system. There are myriad challenges in this subproblem, which derive from the adversary's learning by moving laterally through the network, on the state of knowledge and lack of knowledge about means by which the adversary can make those movements, on quantifying the difficulty of exploiting vulnerabilities as that lateral movement is made. This talk highlights the challenges as we see them after working on this and related problems for several years.