This paper introduces a distributed control architecture for the risk management of hazardous industrial facilities. The objective is to guarantee a high level of risk management in the operating conditions of an industrial facility by means of anticipation, prevention, avoidance and warning of any critical situation that could affect goods and people. Such a situation could appear by any abnormal positioning, handling and storage conditions of physical assets (e.g., pallets with potentially dangerous products, …) as well as their interactions with other assets or moving entities (e.g., forklift, autonomous ground vehicle, …), including human operators. The proposed methodology includes three levels of risk management defined according to different decision time horizons. At the reactive behavior level, static and moving assets (considered as agents) are monitored in order to check that no risky situation occurs. At the proactive behavior layer, a potentially risky situation is avoided by a mobile asset by adjusting its trajectory with only local knowledge of its neighboring assets (e.g., two autonomous forklifts carrying incompatible assets that move to a crossing area) in a distributed way. At the higher level of forecasting behavior, by taking into account the experience from the history and past situations, the warehouse management system can modify the storage locations of certain physical assets.