OpenAI board pick Paul Christiano warns of AI catastrophe risk

Christiano’s appointment pulls one of the most influential alignment researchers back into OpenAI’s power structure at exactly the moment frontier labs are pushing into Astra-class systems. His public statement that the industry is not on track to reduce catastrophic loss-of-control risk, and that most people could die if we fail, moves arguments that used to live in alignment papers and blogs into mainstream political and investor discourse.

For the race to AGI, this is a signal that safety governance is no longer a side show. OpenAI is effectively acknowledging that its own risk profile has shifted enough to warrant putting a doomer-adjacent voice inside its highest safety body. That raises the odds of internal friction over deployment speed, model access and autonomy, especially around agentic coding and computer-use features where Astra excels. It also arms regulators and skeptical politicians with a credible insider who says the current trajectory is not acceptable.

Competitively, Anthropic now has a mirror-image narrative: former staffers publicly warning of extinction risk, and Christiano choosing to exert leverage from OpenAI’s side. Other labs like Google DeepMind and Meta may come under pressure to demonstrate similarly heavyweight safety oversight or risk being painted as the reckless ones. The strategic question is whether these moves actually slow capability scaling or simply provide political cover for it.