Didn't read through the whole thing yet, but this seems to be the key idea: "With LLM-integrated applications, adversaries could control the LLM, without direc

Didn't read through the whole thing yet, but this seems to be the key idea: "With LLM-integrated applications, adversaries could control the LLM, without direct access, by indirectly injecting it with prompts placed within sources retrieved at inference time."