1 paper
Zac Garby, Andrew D. Gordon, David Sands
Large language models are increasingly deployed as agents: they plan, call tools, read untrusted data, and act on the results. This exposes them to prompt injection: data meant onl…