New Framework Quantifies Residual Risk in Agentic AI Systems

Researchers have introduced CPSAINT and FRIESA-K, a novel compositional framework designed to provide a more comprehensive understanding and quantification of residual risk in agentic AI systems. Current risk models often offer only partial views, either describing failure mechanisms without quantifiable risk estimates or providing estimates without detailing the internal failure paths. This new framework bridges that gap by coupling these two perspectives. CPSAINT provides a seven-layer integrity decomposition, spanning physical state, sensors, data, compute, actuators, environment, and time, allowing for a granular analysis of potential failure points. FRIESA-K then maps each identified failure path to a quantified risk instance, grounding the resistance term in a controlled absorbing Markov model. This approach derives control effectiveness from state dynamics rather than informal scores, resulting in a concise mechanism-to-magnitude pipeline for resilient agentic and embodied AI, applicable across diverse scenarios like warehouse robots and financial agents.