Finding a vulnerability in a black box is drastically different from finding one in a white box. This isn’t about whether there is a vulnerability or not, but about the likelihood of it being found.
LLMs, and tools built to use them, are violating a lot of assumptions these days.