Return to Article Details Misuse of large language models: Exploiting weaknesses for target-specific outputs