Anthropic's Claude Mythos found a 27-year-old bug in one of the most secure operating systems ever built and a 16-year-old vulnerability in video software that conventional tools had examined five million times. A researcher learned about the successful exploit via email while eating lunch outside. Mythos is Anthropic's largest model at roughly 10 trillion parameters, six times the size of any previous frontier model, deployed under ASL-3 standards to more than forty organizations through a gated program called Project Glasswing.
Anthropic's red team report states these security capabilities were not explicitly trained. They emerged as a byproduct of optimizing for code, reasoning, and autonomy. The same improvements that make Mythos better at patching vulnerabilities make it better at exploiting them. Security analysis was collateral output. That distinction, between hardening and building, will not hold as access expands.
The structural consequences are concrete. Organizations with access can find zero-days competitors cannot. Any system not scanned at this level is porous by default. Engineering budgets will shift toward hardening. Buyers will start demanding it. The original piece traces exactly how pricing power, security posture, and access hierarchies rewrite competitive dynamics across every company shipping code. That argument is worth reading in full.
[READ ORIGINAL →]