We Are Still Unable to Secure LLMs from Malicious Inputs