New hack uses prompt injection to corrupt Gemini’s long-term memory There's yet another way to inject malicious prompts into chatbots.