Artificial Intelligence systems are increasingly becoming central components of modern IT infrastructures. In this scenario, resilience, understood as the ability to withstand, adapt, and recover from adverse circumstances, has emerged as a core necessity at organizational and societal levels, particularly as errors or failures in AI-driven systems can lead to significant operational, economic, or security consequences. The specific characteristics of AI-based systems, however, introduce novel challenges that require dedicated approaches to guarantee robustness and operational continuity. Of all the AI technologies, Large Language Models have rapidly become the most integral. Their remarkable ability to process enormous volumes of information and support decision-making across a wide range of tasks, including in critical domains and sectors, is contrasted by known vulnerabilities such as hallucinations, susceptibility to adversarial manipulation, degradation under distribution shifts, and difficulties in maintaining long-term coherence and factual accuracy. This article reviews how integrating Knowledge Graphs can significantly enhance the robustness of LLMs. Knowledge graphs provide structured, explicit, and verifiable semantic information. They help mitigate hallucinations by offering a continuously updatable source of truth without requiring full model retraining and improve robustness against incomplete or deceptive inputs. Moreover, the explicit semantic structure of KGs improves explainability, enabling more transparent and interpretable LLM outputs, thereby enhancing both user trust and system-level accountability.