As usual, @dangoodin has written an excellent security explainer article. This one is about prompt injection...but not the usual trial and error whack-a-mole prompt manipulation by pizza guy...instead, automated manipulation by search in gradient space.
This technique is new enough that we're discussing the original paper only today at BIML. It makes the whole boring front door malicious input thing much more interesting.
BTW, the first version of this kind of attack is described in this paper