OpenAI Says It Needs to Expand Disclosure of Alignment Failures

Summary

OpenAI stated that as model capabilities enter a new phase, it needs to expand the scope of disclosure for alignment failures. The industry lacks clear standards for reporting alignment issues during training, evaluation, and deployment, and the company is building a framework and collaborating with dozens of global regulators.