A specialized AI model trained to identify and classify unsafe, harmful, or policy-violating content rather than generate general responses.