AI SECURITY HACKED: We Found the UNSEEN Weakness!

Ajouté : 2026-05-17

[00:00:00]I was the only first one in the world that you could prove that bias was structural in in LM systems. But when I start doing this security layer and I this is strange. I mean, super deep I actually did it because it's so strange. Yeah. We are hearing that Anthropic's new model Methus can identify and exploit security holes in hours. Task that used to take years, for example. Does our current output based security even stand a chance against a model designed to weaponize vulnerabilities so quickly?

[00:00:32]Yeah, it can find holes, but but it the holes need to be there first of all.

[00:00:35]They it cannot create holes. That's it.

[00:00:38]The problem is with with Methus is that it is so capable that if it gets to into the other side behind the firewall and now it has access to the language models, but it is probably good enough to actually hide within the language model.

[00:00:53]And you you can't you can't see it. You can't feel it. You can't do anything because the whole idea is a lot of info in there. So, the only one who has cracked how to to do that is is is us.