Select a category to explore research frontiers
Loading categories...
Doctoral work studies attacks that embed directives inside content an agent processes, and defences against them. This class of vulnerability is intrinsic to agents that read untrusted material.