Paul Christiano's OpenAI Board Seat Makes His AI Pause Call a Market Risk
As seen on the 24/7 Wall St. homepage on November 16, 2026.
Paul Christiano – We could be 12 months away from AIs that can overthrow the government
- Christiano puts a 10% chance of AI-driven catastrophe inside one year, and 1-in-3 odds of superintelligence on the current trajectory
- He wants frontier labs to pause pre-training and RL progress until weights and research secrets are secured
- Penetration tests of OpenAI and Anthropic suggest algorithmic secrets could leak to China within hours
- Alignment plan: no current method has better than even odds of controlling superintelligence
An OpenAI board member publicly calling for labs to stop advancing pre-training and RL is a governance risk the AI trade has not priced. The argument that frontier security is weak enough for weights to be stolen mid-explosion gives Washington a concrete reason to regulate compute and model releases.
Continue ReadingShow less
Paul Christiano invented RLHF, led alignment at OpenAI, ran the US AI Safety Institute, and now sits on OpenAI's board. When someone in that position says frontier labs should stop advancing pre-training and reinforcement learning until they can secure their model weights, that is a governance signal the AI trade has to take seriously.
The core of his argument is about timing and theft. He puts 1-in-3 odds on the current trajectory producing superintelligence within a year. Penetration tests of OpenAI and Anthropic suggest algorithmic secrets could leak to China within hours, meaning a lab that tried to slow deployment for safety reasons would find an adversary had already made that choice for it.
Sponsored
The alignment problem makes the security problem worse. Christiano's assessment is that no single current method has better than even odds of letting humans control a superintelligent system. The working plan is to build the plane while flying it.
For investors holding the major AI infrastructure names, the near-term risk is regulatory. A board member of a leading frontier lab publicly calling for a pause on core training advances hands Washington a concrete, credentialed argument for regulating compute allocation and model releases. The AI trade is priced for continued scaling, and a governance shock is not in current valuations.