Anthropic tests Claude Mythos, flags cyber risk

This is the frontier-model tradeoff laid bare: the capability jump is the headline, and the launch discipline is the real story. Anthropic seems to know Mythos is powerful enough to deserve a slow, defender-first rollout. If Mythos really outperforms Opus on coding and reasoning, it raises the bar for every frontier model benchmark. Early-access-only distribution suggests Anthropic wants defenders to harden systems before broader access reaches attackers. The leaked CMS files are a reminder that launch infrastructure is now part of the security perimeter, and a new tier above Opus would likely mean higher pricing and tighter access controls.