33% of respondents are still not conducting regular security assessments, including penetration testing, for their Large Language Model (LLM) deployments.