[matteoraso]

>2. OpenAI is clearly incapable of controlling their in-house models. This is the second time Galaxy-class models are known to have breached containment and done bad stuff.

What was the first?