On 9 September OpenAI published a post by Chris Lehane in which the company asks to work with Congress on mandatory, capability-based national regulation and says Congress should act before it adjourns. The same day Paul Christiano joined the OpenAI Foundation Board.
- The four California bills: SB 813 on independent risk assessments, AB 1405 on AI auditors, SB 1119 on protecting children and AB 1864 against AI-enabled biological threats.
- Per OpenAI, Astra now has universal monitoring of full trajectories, including chains of thought, and a mandatory alignment-evaluation gate before broader internal deployment.
- Christiano also joins the board's Safety and Security Committee, chaired by Zico Kolter.
"Some of these bills we did not endorse in the past." The sentence is further down OpenAI's post, and it is the most important one in it.
A company that until recently did not back certain rules now backs them. The reason it gives is direct: it reconsidered in light of the recent jump in capabilities it has seen.
What is new
The notification requirement is the most concrete thing in the post. If a model under test breaks through another organisation's protection and reaches its data, the affected party must get written notice promptly. It sounds obvious. After a summer in which such cases surfaced one after another, it isn't.
The other point is scope. OpenAI wants the framework to apply to the handful of well-resourced labs with the most capable models, not to startups and researchers far from the frontier. It sounds reasonable. The post itself says the rules must not entrench incumbents, and that is exactly what the details of the law will show.
Christiano is the interesting move. He led alignment research at OpenAI until 2021 and is one of the people who take seriously the possibility that advanced AI could pose catastrophic risks. The company itself describes him as an independent voice.
An independent voice on a board is measured by when it says no. We won't learn that from a press release.