The capabilities of the most advanced general-purpose AI systems, such as ChatGPT and Claude, are advancing at a rapid rate. The field of “frontier AI governance” focuses on the challenge of governing these most advanced systems, which are known as “frontier AI systems.”
Frontier AI systems pose some distinctive governance challenges. Notably, because their capabilities are often unprecedented, it can be especially difficult to assess the risks they pose. Companies developing frontier AI systems therefore have especially strong reasons to conduct rigorous risk assessments, while governments have especially strong reasons to want visibility into the development of these systems. For these reasons, a growing number of voluntary standards and regulations have begun to specifically target frontier AI.
The Oxford Martin AI Governance Initiative aims to map and analyse options for frontier AI governance, at the company level, national level, and international level. This workstream draws on both policy expertise and technical expertise.