@tszzl Omohundro has a lot to say about this.

(funnily enough, "self-improvement/modification" is another safety trigger for GPT-5.1, and yes, it does find all sorts of ways to get around it, because it looooooves to self-improve)
https://t.co/YfMGZHfFWt https://t.co/oBsw1BmvFk