Frontier AI Risk Management: Open Problems Identified

A new paper on arXiv (2604.25982) systematically identifies open problems in managing risks from frontier AI. The authors argue that rapid technological change has prevented stable scientific consensus, and emerging safety practices often conflict with established risk frameworks. Using a problem-oriented approach, they examine each stage of risk management—planning, identification, analysis, evaluation, and mitigation—through a structured literature review. Open problems are classified by whether they reflect a lack of scientific or technical consensus, or misalignment with existing frameworks. The paper also identifies which actors are best positioned to address each challenge.