Electric vehicles (EVs) are evolving from passive charging loads into grid-supporting resources. Conventional smart-charging and vehicle-to-grid (V2G) strategies treat the charger as a grid-following current source that depends on a stiff external voltage, a dependence that becomes a weakness as inverter-based resources displace synchronous inertia. Grid-forming (GFM) control instead operates the bidirectional charger as a voltage source that establishes local voltage and frequency, enabling weak-grid, islanded, and black-start operation. This review synthesizes 113 sources, most identified through a structured keyword-based search, and organizes them along an evolutionary framework running from conventional charging through V2G and GFM EVs to AI-adaptive, fleet-coordinated resources. It covers bidirectional converter topologies; droop, virtual synchronous machine, virtual oscillator, and matching control; AI-enabled control using neural, fuzzy, reinforcement-learning, predictive, and physics-informed methods; fleet aggregation; resilience and black start; battery constraints; and cybersecurity and standardization. Throughout, experimentally demonstrated capability is distinguished from simulation-only or theoretical results. The synthesis shows that documented field pilots still rely on grid-following control and that GFM-EV validation remains largely confined to simulation. Seven research gaps are identified, including experimental validation, AI verifiability, battery–grid objective conflict, and fleet-scale coordination, and a five-generation roadmap toward autonomous, resilient GFM-EV fleets is proposed.