The published transcripts are the most valuable part of this. We've found that real exploit chains almost never look like what you'd dream up internally. One thing I'd push on is are the agents stateful across attempts?…
The part people miss is that prompt injection is just the delivery mechanism but the actual vulnerability is that there's no enforcement layer between the agent's decision and the action firing. You can harden the…
The published transcripts are the most valuable part of this. We've found that real exploit chains almost never look like what you'd dream up internally. One thing I'd push on is are the agents stateful across attempts?…
The part people miss is that prompt injection is just the delivery mechanism but the actual vulnerability is that there's no enforcement layer between the agent's decision and the action firing. You can harden the…