Cascade Poisoning

A security threat in multi-agent systems in which untrusted input contaminates an upstream agent, and its structured output is then accepted by downstream peer agents as a verified factual baseline, progressively amplifying the security breach through the workflow until it escalates into unauthorized privileged operations.