AML.T0061

MITRE ATLAS / LLM Prompt Self-Replication

An adversary crafts a prompt injection designed to make the model reproduce the injected prompt in its own output, so the instruction spreads into subsequent contexts, documents, and conversations.

Think of it likeA chain letter that instructs each reader to copy it out again before passing it on.

In plain English

An injected instruction that tells the model to repeat itself will end up in summaries, replies, and saved notes. From there it gets read by the next model and the infection keeps going.