An attacker specifically designs prompts that are too long for the LLM's context window, intending to overwhelm its memory and inadvertently expose confidential...