No, they admitted that too, that they made a harness to do it. And Jake said he was trying it “similar attacks” on others too and then later the head guy said it’s actually “only primcloud” and refused to comment on it anymore after that.
And like you said, this is not some amateurs who makes such mistakes, it’s company serving 5M people, and I am sure they hire competent people at that scale.
And no model is going to do what he did unless he is using an abliterated model.
It is literally a crime and there is no denying it.
I already found it weird that no gov/org is holding OpenAI accountable for all the unsolicited hacks they keep on doing.
And it’s always “oh the agent went out of control” that’s bullshit..
Jake literally admitted to it? And the primcloud guy posted logs showing all the requests? What else do you need?
They admitted everything already, there is no need to “think” anymore.
They announced that they specifically left out any kind of cyber security related training, but i guess it can still work tho. Or a LoRA on top will fix it.
Damn..everyone in the comment is a paid shill. But who they working for? Which marketing agency is this 😂
