Inside the scramble for trusted AI cops

The race to regulate AI and ensure its safe operation is heating up, with industry leaders, policy makers, and businesses scrambling to establish trustworthy standards. At the White House, progress has been minimal, with a general consensus that swift federal action is unlikely in the near future. AI companies are taking matters into their own hands, grappling with how to test their advanced systems and work with China on safety measures.

There is a growing effort to identify third-party evaluators of AI systems, but concerns have arisen over potential conflicts of interest among current evaluators, who are seen as too closely tied to the top AI companies. Some suggest alternative options, including businesses, startups, and even individuals from within the AI research community.

However, the lack of a unified approach and the close-knit nature of the AI research community have led to skepticism about the effectiveness of these proposed solutions. Despite these challenges, transparency through voluntary incident reporting, such as OpenAI's public incident reporting playbook, is seen as a step in the right direction, although some argue that it may not be sufficient without more robust oversight.