Abstract Unmanned aerial vehicles (UAVs) are now used in logistics, agriculture, surveillance and the low-altitude economy, but their growth has created two linked problems: keeping UAVs themselves safe from faults and attacks, and protecting airspace from unauthorized UAVs. This paper reviews 111 studies compiled from seven source papers published between 2019 and 2026 (two comprehensive surveys on counter-UAV sensing and flight-data anomaly detection, one survey on machine learning and large language models (LLMs) for UAV safety and security, and four technical studies). The studies are organized into five areas: (i) multi-sensor detection, identification and localization of hostile UAVs; (ii) physical and cyber countermeasures; (iii) flight-data anomaly detection and recovery; (iv) sensor-malfunction and GNSS-spoofing detection; and (v) vibration-based fault detection and AI-driven safety and security, including LLMs. For each area we summarize the methods, reported accuracy and stated limitations, and compare them in tables. The review shows that deep learning with sensor fusion consistently outperforms single-sensor and classical methods, but that results are hard to compare because datasets differ, anomalies are usually injected artificially, and real-flight validation is rare. We close with a set of research gaps and future directions: public benchmarks, real-flight testing, onboard-efficient models, federated and LLM-based approaches, and joint safety-security frameworks.