Christina Criddle

That means that they're left vulnerable to attack by hackers and potentially foreign adversaries.

Yeah, I mean, I would just say there is some skepticism about how elevated Claude Mythos is with cybersecurity compared to what we've seen before.

But I definitely think this is a moment where wider sectors and industry and governments are recognizing this as a problem.

And then to add on the pressure here of the threat, Anthropic is investigating this report, which claims that a group of people had unauthorized access to Claude Mythos.

Apparently, they were able to access the model through a third-party vendor environment, as one member of the group apparently was a contractor for Anthropic.

And in the run-up to Mythos, there were already two security issues.

So there is this criticism that maybe Anthropic doesn't have its own security and order.

And what does that then mean if unauthorized people are able to access this model, which they claim is so powerful?

And if this gets into the wrong hands, that could be really dangerous.

Well, what has Anthropic said about these issues?

Anthropica said that they're investigating this report, which claims unauthorized access.

They say it's through one of this third-party vendor environments and not through its API, which is how most developers would access this model.

And in terms of the previous two security issues in the run-up to Mythos, they've said this was human error, this was not caused by AI, but that still raises quite a big question about how secure their infrastructure is.