Responsible reporting for frontier AI development

Noam Kolt, Markus Anderljung and other researchers propose a responsible reporting framework for frontier AI development, arguing that organizations that develop and deploy frontier systems should report safety-critical information to governments, businesses and civil society to improve visibility into emerging risks. The paper argues that with this information, developers can make better risk management decisions, and policymakers can design more targeted and more robust regulatory infrastructure. The paper outlines the key features of responsible reporting and proposes mechanisms for implementing these features in practice.