We propose TripWire as a framework that provides for the mediation of software self-awareness by supporting real-time assessment and response capabilities. Our approach is inspired by the recent success of automatic speech recognition systems, which can assess the likelihood of a potentially unbounded set of possible utterances and select the most likely candidate in real-time, given an underlying model of the conversational domain. We see an analogy between estimating an utterance from an acoustic waveform and estimating the behavior of a program from dynamic-analysis data. In both cases, there is an inherently stochastic causal relationship between the quantity of interest and the measurement data. Our strategy is to leverage the successful tools and methods of speech recognition in the novel context of program behavior analysis. TripWire technology provides a systematic approach to behavior modeling and recognition with three main elements: strong domain knowledge and tools, learned statistical models, and real-time classification processes.