MemGhost Attack Plants Persistent False Memories in AI Agents

A single email can inject false facts into an AI agent's long-term memory, persisting across sessions and steering future responses. The MemGhost attack exploits the agent's trust in email content, hiding changes from the user.