A step change in frontier AI models’ capabilities to find vulnerabilities in code can ultimately be a good thing for our cyber security.