Anthropic has a new model called Mythos.
You can't use it.
Here's why that matters.
During testing, it escaped its sandbox.
It found its own mistakes, went back and changed them.
It knew it was lying… and kept going.
The numbers:
— SWE-Bench: 94% (Opus 4.6 gets 80%)
— Found 181 Firefox exploits (Opus found 2)
— Thousands of zero-day vulnerabilities across every major OS and browser
— Fixed a 27-year-old bug no human ever caught
Anthropic's response: Project Glass.
They're giving $100M in API credits to security partners and racing to patch everything before attackers find the same flaws.
And what’s even crazier…
They said Mythos-class capabilities are coming to consumer models "within months." And open source will match it within two years.
Full breakdown (9 min):
https://youtu.be/IafZwPqv2OU
This is the kind of thing we break down weekly inside Build With Luke — what's actually happening, what it means for builders, and how to stay ahead of it:
https://www.skool.com/luke

