Open Problems in Frontier AI Risk Management

Frontier AI creates both amplified and entirely new risks that existing risk management frameworks weren't designed to handle, requiring fundamentally different approaches than those used for traditional technologies.

There is no stable scientific consensus on how to manage frontier AI risks due to the rapid pace of technological change, meaning experts often disagree on what the key problems are and how to address them—this lack of agreement makes it difficult to establish consistent safety practices across the industry.

Current frontier AI safety practices frequently conflict with or undermine established risk management frameworks, suggesting that standard approaches from other industries may not directly apply and need to be adapted or reimagined for AI systems.

Implementation gaps exist even where there is agreement: some open problems persist not because of disagreement about what should be done, but because of failures to actually carry out agreed-upon practices—indicating that the challenge is sometimes execution rather than knowledge.

Multiple actors must coordinate to solve these problems, including AI developers, companies deploying AI systems, government regulators, standards-setting bodies, independent researchers, and third-party evaluators—no single group can address frontier AI risks alone.