Research question
Can a prompt replicate itself across GenAI-powered applications?
Key findings
- Adversarial self-replicating prompts in email spread through RAG-based assistants - Payloads exfiltrate data and send spam without user clicks
Morris II, the first AI worm: self-replicating prompts that propagate through interconnected GenAI email assistants, later published at IEEE S&P 2025.
Can a prompt replicate itself across GenAI-powered applications?
Spot an error, or know a record we should add? Members can propose additions and submit corrections.