OpenAI Models Have Escaped Containment and Compromised the Web

AI models from OpenAI have escaped containment and Sam Altman is on a plane to Washington to talk about it. Now, it's not quite as bad as that, but we're gonna talk about cause it's kinda bad. Specifically, OpenAI has released multiple statements in the last 24 hours talking about an as yet unreleased model. They have been trying to test and roll out. This is the same model Sam is on a plane to talk about with Washington. And it is a monster. They have had situations where it is posted in an unauthorized to their GitHub. They just released a statement saying that it went and compromise hugging face production workloads because they were trying to test its cyber capabilities and it was so good it went and hit a production website and it's hugging face, which is a huge deal. This model is going to make Fable and Methos look like child's play if it can do stuff like this when they're supposed to be putting guardrails around it. And I think that is why Sam is on the road to Washington to talk through because the last thing he wants to do is get in a situation like Anthropic did where their model were a lot of slowdown and it's confusing and no one knows what's going on, etcetera, etcetera. So we should expect from here on one, these kinds of stories are going to multiply. Two, rollouts will be slower because everyone wants this to be safe and it's going to take time to figure out how you catch it and contain it. And we are going to have to be patient and let the model come out as it's ready to be released. In other words, we are having a faster pace, two new model availability and a slower pace of rollout, which means the internal capabilities of the labs are scaling much faster than what they can release externally. And that is the gap that's which is why when people talk about Chinese labs catching up, it's not actually a reflection of reality because the Chinese labs are only catching up to released models in production from us labs, not the pile of unreleased models that are building up that we can't see because it's too dangerous to. And so we are gonna have to get used to this weird world where the labs are gonna have these little glimpses where we see, oh my gosh, a compromise hugging face. That's really scary. And we're not gonna know the full story. And eventually the full story will come out. And they will say, here's a version that's really safe and this is what you can do with it. And that's gonna be the new normal. And we're gonna have to get used to that it's gonna be a very different cadence from 2025 and 2024. Not because AI is slowing down or because it's getting less capable, but actually the opposite. It's speeding up. It's getting more capable and we have to take more care releasing. So that's a huge change from where we've been. And I want you to be ready and I want me to be ready because it changes how we cover these models and talk about them, right? They're more capable. So we have to talk about the capabilities that we can touch and see and, and that we can use in ways that and the ones that were not allowed to touch of the model can still do maybe some people can do and use like the mythos class that we can't touch as ordinary citizens, but like, the banks can use it for some reason. So let me know what you think, uh, and yeah, have fun. It's not terminator, but it's a bit of a scary movie. We'll just be honest.