Nothing you're bringing up matters. OpenAI is the creator and operator. They're legally culpable for the consequences of the machine they made. The model is a machine: even if it could be demonstrated that the model reasoned its way into criminal behavior completely independently of OpenAI staff, that doesn't change anything.
If I run a biology lab and engineer a terrible virus, it gets out, and a global pandemic ensues, I don't get to shrug and say "well we told it not to infect people". It's my fault for failing to mitigate the risks of my work.
If I run a biology lab and engineer a terrible virus, it gets out, and a global pandemic ensues, I don't get to shrug and say "well we told it not to infect people". It's my fault for failing to mitigate the risks of my work.