Turkish Air Force Academy

Recent Advances and Open Challenges in Mitigating Inference-Time Attacks on Large Language Models

The rapid integration of Large Language Models into high-stakes domains has elevated inference-time attacks into a primary security concern for production deployments. These attacks are adversarial techniques that exploit models exclusively through their input–output interface. T...