Fighting AI with AI: AI-Agent Augmented DNS Blocking of LLM Services during Student Evaluations

Researchers introduce AI-Sinkhole, an AI-agent augmented DNS-blocking framework that dynamically detects and temporarily blocks LLM chatbot services during proctored exams to prevent academic integrity violations. The system uses quantized LLMs for semantic classification and Pi-Hole for network-wide DNS blocking, achieving robust cross-lingual detection with F1-scores exceeding 0.83.

AI-Sinkhole addresses a growing tension in education: leveraging AI's transformative potential for accessibility and personalized learning while preventing its misuse during assessments. Academic institutions face unprecedented challenges as LLMs enable students to bypass critical thinking requirements, shifting cognitive load away from analytical skills. The framework represents a defensive technological approach where AI combats AI, using machine learning classifiers to identify emerging chatbot services before blocking them at the DNS level during examination periods.

This development reflects broader institutional concerns about authentication and verification in the AI era. Universities have struggled with ad-hoc solutions—from honor codes to plagiarism detection tools—each proving insufficient against LLMs' sophistication. AI-Sinkhole's deployment of quantized models (Llama 3, DeepSeek-R1, Qwen-3) demonstrates practical resource constraints in educational settings while maintaining competitive performance. The open-source availability signals a shift toward collaborative defense mechanisms across institutions.

The system's significance extends beyond education into organizational security and access control paradigms. As LLM services proliferate, DNS-based blocking becomes a viable network management strategy, though it raises questions about blanket restrictions versus granular access policies. The cross-lingual performance metrics suggest the approach scales internationally, addressing global examination integrity challenges. Organizations implementing similar architectures could extend this model to controlled environments beyond academics, including corporate training and certification programs.

• →AI-Sinkhole uses quantized LLMs to semantically classify and block emerging chatbot services during proctored exams with F1-scores exceeding 0.83
• →The framework combines AI-agent classification with Pi-Hole DNS blocking for dynamic, network-wide enforcement during assessments
• →Cross-lingual detection capabilities enable global institutional deployment across diverse linguistic contexts
• →Open-source code release accelerates adoption of defensive AI mechanisms across educational institutions
• →The approach demonstrates practical implementation of AI-versus-AI security paradigms in controlled institutional environments