Yes, this is one specific safety problem -- there are many other RL safety problems that deserve high quality benchmarks too. See eg https://arxiv.org/pdf/1606.

Yes, this is one specific safety problem -- there are many other RL safety problems that deserve high quality benchmarks too. See eg https://arxiv.org/pdf/1606.06565.pdf https://arxiv.org/pdf/1606.06565.pdf or https://medium.com/@deepmindsafetyresearch/building-safe-artificial-intelligence-52f5f75058f1 https://medium.com/@deepmindsafetyresearch/building-safe-art... for discussions of the problem space.