A Call for Open Science in AI Safety (make-safety-open.github.io)

A coalition of AI researchers, led by notable figures from institutions like EPFL and Oxford, has initiated a petition advocating for greater transparency in AI safety practices among frontier model developers. This call emphasizes the necessity for developers to disclose detailed methodologies and evaluation results that can be independently reviewed. By fostering an environment of openness, the initiative aims to facilitate scientific scrutiny and accelerate advancements in understanding and improving AI safety mechanisms. The significance of this movement lies in its potential to harmonize safety research with the rapid evolution of AI technologies. By sharing critical safety-related information—such as training protocols, evaluation metrics, and documented behavioral outcomes—researchers can better assess the safety of AI models and target funding towards the most promising safety enhancements. While exceptions to information sharing are acknowledged in cases of potential misuse, the overarching goal is to utilize collective knowledge within the global scientific community to enhance the safety and reliability of AI systems, ultimately ensuring a more responsible deployment of powerful AI technologies.