Palo Alto Networks used Anthropic's Mythos Preview and OpenAI's GPT-5.5-Cyber to audit its own product codebase, finding 75 legitimate vulnerabilities across more than 130 products in approximately one month — over seven times its typical discovery rate of 5–10 bugs per month. The models generated working exploits more than 70% of the time and were particularly effective at chaining multiple low-severity weaknesses into high-severity exploit paths. Palo Alto estimates organizations have a three-to-five-month window before threat actors gain equivalent access to frontier AI cyber capabilities.