Recent research on intrusion detection mechanisms in MANETs emphasises the transition from traditional signature-based systems to adaptive, intelligent IDS frameworks that operate under the dynamic, infrastructure-less conditions of mobile ad hoc networks. These frameworks commonly integrate machine learning and deep learning models (e.g., ANNs, CNN-LSTM, enhanced Naive Bayes) to classify normal vs malicious behaviours even when topology changes rapidly or resource constraints are strict. Feature-selection methods such as canonical correlation analysis and heuristics like the dragonfly algorithm are used to reduce dimensionality and improve detection speed, while fuzzy cognitive adaptive systems handle uncertain and evolving attack patterns. Experimental results report detection accuracies in the range of 90-96% and significant reductions in false positives (<5 %), demonstrating viability in real-time mobile environments. At the same time, research acknowledges challenges including computational overhead on mobile nodes, adaptability to unseen zero-day attacks, and balancing detection performance against energy consumption and mobility. Overall, the field is moving toward lightweight, learning-based, mobility-aware IDS designs tailored for the unique constraints of MANETs.