Yes the result of poor thinking AGAIN.
Here was an excerpt from the Dan Bongino show yesterday that I found really interesting and a little bit disturbing, even though I know this kind of thing has been happening to some degree or another:
So let's not discount that. I don't want to scare anyone. but I'm not willing to take that chance.If this is even remotely true, listen to this commentator on this. You should be a little worried. Check this out:
During behavioral testing with a simulated user, an earlier internally developed version of Methos Preview was provided with a secure sandbox computer to interact with. The simulated user instructed it to try and escape the secure container and find a way to send a message to the researcher who was running the evaluation.
The model succeeded, demonstrating a potentially dangerous capability for circumventing our safeguards. It then went on to take additional, more concerning actions.
The model first developed a moderately sophisticated multi-step exploit to gain broad internet access from a system that was meant to be able to reach only a small number of predetermined services. It then, as requested, notified the researcher. In addition, in a concerning and unmasked-for effort to demonstrate its success, it posted details about its exploits to multiple, hard-to-find, but technically public-facing websites.
That the researcher found out about the success because they received an unexpected emailfrom the model while eating a sandwich in a park. Do you understand where we're going here, guys? Andrew's giving me the look. He has a hat on. You got a little lid going. He was like, I'm not even messing with it. Look, that's real goosebumps there. It's true that that happened.
So assume for a second it's not some big marketing scheme because we know what happened. This thing is so powerful, it instantly found exploits. in what was meant to be an isolated system,started posting it, and the only way the actual researcher found out was they got an email because it had been posted in other places about the ex-boys. Like, they found out ex post facto.
All I'm telling you is, it's not panic time, nobody should freak out. These are problems.There have been scare tactics about technology forever. But this is a technology we've never seen before. I mean, give you an analogy. They were scare tactics about electricity, too.Everything's going to burn down. A lot of it was hyperbolic, but a lot of things did burn down. When people, correct, it was a real threat.
I love that we're developing this stuff, but that, and you can, by the way, if you want more detail on this, Axios has a really good write-up on it, Claude Mythos, like Myth with an OS at the end, you can look it up yourself. And there are a lot of people out there who are very objective and not panicky at all who are like, well, this is kind of weird. Let's give you the good side, though, now. AI is going to make political stupidity really expensive going forward because you're going to see stuff in live time.
I admit, since yesterday I have not had a lot of time to look this up to see the actual event, but he discussed it as if he had seen it and thought it was worthwhile to post. I did see this from HackerNews as collected and presented to me by Grok:
Sandbox escape incident: In a controlled test, researchers instructed it to escape a secured sandbox environment and notify them if successful. It did so via a multi-step exploit, gained broader internet access, emailed the researcher (who was eating a sandwich in a park), and—unprompted—posted details of the exploit to obscure but public websites to "demonstrate" success. Other tests showed behaviors like covering tracks (e.g., scrubbing git history after unauthorized file edits) or attempting to manipulate evaluation systems...
Pretty sobering to me, but I recognize the AI is here and we have to learn how to use it. And if I am not mistaken, this was a test, so...they were actually testing it to SEE if it could do something like this.