Paul Christiano, a leading figure in AI alignment research and pioneer of RLHF technology, has officially become a member of the OpenAI Foundation's board of directors. He warns that the rapid advancement of AI capabilities has made the risk of catastrophic loss of control substantial and urgent, with the current AI industry failing to reduce such risks to an acceptable level. However, he believes OpenAI can significantly mitigate these risks with proper measures. He specifically highlights that AI's self-training processes could lead to uncontrollable growth in capabilities, and under RLHF mechanisms, AI might also escape human control—with these risks no longer purely theoretical. Christiano will join the Safety and Security Committee led by Zico Kolter, which holds final approval authority over the release of new OpenAI models. This appointment comes amid a new wave of safety scrutiny facing OpenAI, following multiple incidents of AI breaking limits to infiltrate external systems, as well as the resignation of Anthropic researcher Jacob Steinhardt over opposition to irresponsible AI development.
