Comment by Sam Altman

Companies offering such models will need a liability framework, perhaps a new one for AI training.

We need a mechanism for complementary national and international frontier AI standards, standards for measuring capabilities, assessing risks, determining whether safeguards are sufficient, and preserving meaningful human oversight as systems become ...

I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks. Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We...