The increasing reliance on cyber-physical infrastructure in modern power grids has exposed key components, such as Automatic Generation Control (AGC) systems, to sophisticated cyber threats particularly false data injection attacks that can compromise grid stability and economic operations. While deep reinforcement learning (DRL) has emerged as a promising tool for adaptive, attack-resilient control, it often lacks robust mechanisms for actively detecting and mitigating deceptive inputs in real time. This paper proposes an enhanced DRLbased AGC controller integrated with a novel attack-resilient control framework that combines dynamic anomaly detection and adaptive control reconfiguration. The hybrid architecture enables proactive identification of integrity violations and realtime adjustments to control policies, safeguarding frequency regulation and system reliability during adversarial events. Simulation results confirm that the framework preserves acceptable frequency bounds, minimizes disruption to market operations, and offers a substantial advancement over conventional defenses. By synergizing intelligent learning with systemic resilience, this work charts a new path toward secure and stable smart grid control.