Cohen, Bitton and Nassi present Morris II, an adversarial self-replicating prompt that propagates through RAG-based GenAI email assistants, causing data exfiltration and further spread. The paper also proposes a detection guardrail and reports its accuracy.
Why it matters
It showed that prompt injection can propagate between connected assistants, a precursor to multi-agent attack concerns.
Key facts
As stated in the sources, with where to find them.
- The proposed Virtual Donkey guardrail is reported at a true-positive rate of 1.0 with a false-positive rate of 0.015.Abstract
- Evaluation used GenAI-powered email assistants across context sizes, embedding algorithms and propagation chains.Abstract
Findings that cite this record
Key questions this bears on
- Can prompt injection against AI agents be reliably defended?Not reliably. Bounding what untrusted input can trigger helps, but trust boundaries must survive delegation and tool execution.
Sources
Related records
Sep 28, 2026
Mar 5, 2024
Apr 30, 2026
Sep 20, 2024
Aug 20, 2024
Nov 3, 2023