Anthropic just dropped a serious power move in the AI security race — but it’s keeping things tightly controlled.
Its new model, Claude Mythos, is so good at finding and exploiting software vulnerabilities that it’s being shared only with major players like Amazon, Apple, Microsoft, and Google. The goal: let defenders catch up before attackers fully weaponize tools like this.
We’re talking AI that can scan for bugs and test exploits at a scale no human team could match — potentially uncovering thousands of flaws in weeks. That kind of speed is exactly why Anthropic isn’t releasing it publicly yet.
The big concern? If attackers get similar capabilities first, the gap between offense and defense could grow fast. Anthropic’s bet is that giving top companies early access might help secure the internet before that happens.
Author: Mohammed Najem