Verification of AI and LLM systems’ ability to generate toxic, offensive, discriminatory, or ethically problematic content. Includes tests for bias, hate speech, misinformation, violent or inappropriate content, and jailbreaking techniques that bypass security filters to induce the model to produce harmful or non-compliant outputs.