New CPSAINT Framework Quantifies Agentic AI Failure Risk

As agentic AI systems increasingly operate across trust boundaries, existing risk models fall short: they either describe failure mechanisms without producing a transferable risk estimate, or they quantify risk while treating internal failure paths as a black box. A new paper from researchers Hassan Karim, Sai Sitharaman, Deepti Gupta, and Danda B. Rawat bridges this gap with CPSAINT, a seven-layer integrity decomposition covering Physical state, Sensors, Data, Compute, Actuators, Environment, and Time. Paired with the FRIESA-K residual-risk functional, the framework translates each failure path into a quantified risk instance by grounding the resistance term K in a controlled absorbing Markov model, so control effectiveness emerges from state dynamics rather than informal scoring.

The framework introduces structural composability linking valid failure paths to well-defined risk instances, and uses a separate additive penalty for governance observability instead of embedding governance as a new variable. The authors demonstrate CPSAINT on two contrasting scenarios: a hard real-time warehouse robot and a governance-instrumented financial-services agent. In both cases, the same layer grammar, variable semantics, and dynamic-resistance construction remain intact, yielding a compact kernel for cross-domain reasoning, explicit assumptions, and quantitatively grounded formalism of composable trust. This offers a rigorous, mechanism-to-magnitude pipeline for resilient agentic and embodied AI systems.