Text this: A Data-centric approach for safe and secure large language models against threatening and toxic content.