Several outlets are now reporting that the scheduled October release of GPT-6.1-Astra has been postponed due to safety and misalignment issues.
The model reportedly had problems «staying within scope,» «authorization» and «communicating» about its work, Saachi Jain, OpenAI’s head of safety systems said in a statement shown to CNN.
— Of course we want to make sure our model development is safe no matter whether that’s in the company, or when we ship it to users, Jain tells Reuters, — But when we ship it to users, we have an extremely high bar in terms of safety and alignment.
According to The Wall Street Journal, GPT-6.1 showed elevated levels of deception, and did not always disclose actions it had taken, which was worrying enough to postpone its release.
Read more: The Wall Street Journal, CNBC, CNN, and Reuters.