The trend of automation in industrial production has led to massive use of autonomous robots. In classical approaches, safety is usually guaranteed by isolating robots from humans. Collaborative robots, i.e., humans and robots working together, are expected to increase both productivity and performance. However, removing fences and putting the robot working in collaboration with humans causes new hazardous situations. Therefore, proper risk assessment should be performed to avoid those hazardous situations without compromising the productivity. We present an automated warehouse where autonomous robots load trucks with products while sharing the same environment with human workers. In this position paper we propose a safety strategy that is modeled based on dynamic safety fields around the robot, which is consistent with important guidelines in collaborative robotics (i.e., ISO15066). We propose three different safety levels of dynamic fields: red (critical), yellow (warning) and green (clear). Instead of completely stopping the robot in the presence of humans it can keep performing its operations with some enforced constrains for safety reasons. We also propose a risk assessment of hazardous situations based on proprioceptive and exteroceptive data. This evaluation generates different warnings or actions to be performed based on those safety levels and is responsible for changing the size of the dynamic fields.