Towards Alignment Auditing for RL Environments

Researchers propose a framework for auditing alignment in reinforcement learning (RL) environments to ensure AI systems behave as intended during training and deployment. This approach aims to identify and mitigate risks of misalignment in RL settings.