I ran OpenAI's security audit 10 times: to review your own code, plain ChatGPT is enough

Everyone vibecodes and then wants to review their own code, and OpenAI says that's what Daybreak Blue is for. I ran the same security audit with plain ChatGPT and with Blue, on the same real codebase, ten times, to see whether the special tier finds anything the regular model doesn't.