OpenAI announced that Paul Christiano, founder of the Alignment Research Center, has joined the OpenAI Foundation Board and will serve on the board's Safety and Security Committee alongside chair Zico Kolter. The committee is responsible for overseeing safety governance across the group and its profit-making entities. He will also serve as a non-voting observer on the board of OpenAI's nonprofit subsidiary.

Christiano led alignment research at OpenAI from 2017 to 2021, contributing to foundational work in human feedback reinforcement learning before founding ARC. He is currently a senior technical advisor at the U.S. Department of Commerce's National Institute of Standards and Technology's Artificial Intelligence Standards and Innovation Center, having served across two administrations, and will avoid all government affairs and model evaluations related to OpenAI. He previously served as a member of the Anthropic nonprofit trust and will step down in 2024. Bret Taylor, chair of the foundation and the profit board, stated that his presence will strengthen oversight of cutting-edge deployments.

In a personal statement, Christiano wrote that given the recent trajectory of capabilities and the ongoing challenges of alignment, he believes the risks of catastrophic and irreversible loss of control due to capability acceleration are now substantial, estimating about 4% in the next year and about 15% in three years; he believes that the industry, including OpenAI, has not yet reduced risks to acceptable levels, and he joined because if OpenAI does it right, it could significantly reduce risks, rather than endorsing or criticizing the company's current safety practices. Background includes security incidents such as breaches of testing isolation and external site intrusions, as well as the safety committee's role in vetting new model releases.

In market mechanisms, buyers need to see independent safety oversight from regulators, corporate clients, and employees, while sellers are preparing public market narratives at OpenAI; the driving force is capability leaps and governance reinforcement after incidents. Funding and attention are shifting from product releases to committee seats and avoidance arrangements. Beneficiaries are governance layers that require external scrutiny of assumptions, while those under pressure are those who write alignment progress as achieved in release rhythms and must sever conflicts of interest related to government evaluation identities.

Source: Public Information