level: research
agentic ai systems are being deployed in high-stakes settings faster than risk models can keep up. current methods either describe failure mechanisms without giving a transferable risk number, or they output a risk score while treating the internal failure path as a black box. this paper introduces cpsaint, a seven-layer integrity decomposition covering physical state, sensors, data, compute, actuators, environment, and time. it is paired with friesa-k, a residual-risk functional that maps each failure path to a quantified risk instance.
friesa-k grounds the resistance term k in a controlled absorbing markov model. this means control effectiveness is derived from state dynamics rather than assigned as an informal score. the approach creates a concise mechanism-to-magnitude pipeline for resilient agentic and embodied ai. the authors also report governance observability through a separate additive penalty, making the framework useful for oversight and accountability.
the work addresses a gap between qualitative failure analysis and quantitative risk assessment. by coupling these two views, cpsaint and friesa-k provide a structured way to estimate residual risk from specific failure paths. this can help developers and regulators understand not just what can go wrong, but how much risk remains after controls are applied. the framework is designed for agentic and embodied ai systems that interact with physical environments.
why it matters: it gives ai practitioners a method to turn failure analysis into concrete risk numbers, supporting safer deployment and regulatory compliance.