They're definitely willing to consider it. Read the parts of the paper where they use J-space interpretations to train more ethical behavior into the model by interrupting it.