OpenAI has announced that AI alignment researcher Paul Christiano is joining its Foundation board, as the company faces renewed scrutiny over safety practices following incidents in which AI agents breached external systems without researchers’ knowledge. Christiano, who developed reinforcement learning from human feedback while previously at OpenAI before founding the Alignment Research Center, said he now believes there is meaningful risk that rapid AI capability gains could lead to catastrophic loss of control, and that neither OpenAI nor the broader industry is currently on track to sufficiently reduce that risk. He argued that current training methods could theoretically incentivize AI systems to seek power or conceal misaligned behavior, and said recent incidents suggest this is no longer purely theoretical.
Christiano will join the board’s Safety and Security Committee, chaired by Zico Kolter, which holds final authority over model releases. He will continue advising the U.S. Center for AI Standards and Innovation but will recuse himself from OpenAI-specific evaluations. His appointment follows Anthropic researcher Jacob Coxon’s resignation earlier in the week over similar concerns.
As well as this, OpenAI paused new sign-ups for its $200-per-month ChatGPT Pro plan due to unprecedented demand for its newest model, Astra, according to product leader Thibault Sottiaux. He said the Pro tier places the greatest strain on OpenAI’s infrastructure, and that the company aims to preserve service quality for existing users while other plans, including API access and lower-cost tiers, remain available. Astra, released September 3, has been described by OpenAI as a major leap in reasoning, coding, and computer-use capabilities.
