A hacker who says their group found a serious OpenAI exploit argues that ordinary cloud services, employee browsers and third-party dependencies expand AI labs’ exposure to attacks. They question whether that setup fits the labs’ own warnings about nuclear-level risks, while stressing that finding one exploit does not mean they know how to secure an organization as complex as OpenAI. Another commenter echoes the concern, arguing that frontier AI companies’ security practices fall short of the stakes they describe.