Turkish-first LLM security · prompt injection · red-teaming · adversarial datasets · OWASP LLM Top 10 · AI safety