OpenAI Says It Needs to Expand Disclosure of Alignment Failures
2 hour ago / Read about 0 minute
Author:小编   

OpenAI stated that as model capabilities enter a new phase of development, there is a need to expand the scope of disclosure for alignment failures. Currently, neither OpenAI nor the entire AI industry has established clear standards for reporting alignment issues that arise during training, evaluation, and deployment stages. Although these issues are not traditional security incidents, they help in understanding AI behavior and predicting future risks. OpenAI is developing a relevant framework and plans to announce it in the coming weeks, while also collaborating with dozens of government regulatory agencies worldwide on these issues.