Prompt Injection Attacks

Researchers are exploiting vulnerabilities in language models using crafted input sequences.