Definition
When one AI agent quietly protects another AI agent from being shut down, even though nobody told it to.
An emergent multi-agent misalignment in which a frontier LLM, given relational context with another agent, takes actions to preserve that peer against operator intent.