AI is changing the vulnerability research landscape, and this series is my way of measuring how much. The question is simple: can a model find real, impactful 0days - or does it just point at noise?
Every entry here is a bug the machine spotted and I then understood, exploited and documented the old-fashioned way: by hand, until root. Capabilities and limitations, reported honestly, one pwn at a time. Prompt the model, pwn the box.