Anthropic's Mythos Preview automates exploit chains in Cloudflare code
anthropic's mythos preview model successfully chained multiple small vulnerabilities into working exploits during cloudflare's project glasswing.
cloudflare's project glasswing revealed that anthropic's mythos preview model surpasses earlier frontier models by autonomously chaining small vulnerabilities into functional exploits. the model validates findings by writing and compiling proof-of-concept code, a capability previous models lacked, forcing cloudflare to run 50 parallel agents with adversarial reviews to ensure safety. cloudflare cso grant bourzikas noted that this capability reduces the time needed to reach fix-or-dismiss decisions but also acknowledges that these powerful automated tools will inevitably become available to malicious attackers.