Comment Re:Read all the details (Score 1) 33
Unfortunately, there's a lot of people (including some here on Slashdot) who have tried to argue that these instances of AIs breaching their containment have been deliberate hype by the AI companies. The level of detail in this instance and the fact that METR and others have looked at this should make it clear that's not what is going on.
No it makes it clear that is exactly what is going on. Look if all it did was find an open proxy or something and go nuts nobody would be impressed at all and everyone would reasonably conclude it was staged/deliberate. Basically they had to put up enough road blocks and make sure the attack chain would contain enough steps that to human who is used to getting bored, tired, lost trying to follow enough steps looks at it and is bamboozled the number of things, without realzing that each thing is trivial enough...
They obviously had no basic egress monitoring, no IPS/IDS sensors, no flagging of unusual traffic like say to competitors, nothing...or they did have those things and everyone studiously avoided looking at or attempting to understand any of the alerts that surely must have fired for days.
Is it hard to understand the haystack of AI reasoning and agent to agent comms? Sure, but detecting this and stopping this required NONE of that! It is difficult to conclude multiple technical persons at OpenAI at the very least reasonably did know or reasonably should have know the systems were behaving in anti-social and likely illegal ways with outside resources, and chose to take no action.