This study presents a deep reinforcement learning (RL) approach to avoid tearing instability in tokamak plasmas, crucial for stable and efficient fusion energy production. Tearing instability, a leading cause of plasma disruptions, is challenging to predict and control, especially in the ITER baseline scenario. The researchers developed a dynamic model to estimate the likelihood of future tearing instability based on diagnostic signals and actuator data. This model was used as a training environment for an RL-based AI controller, enabling automated prevention of tearing instabilities. The AI controller was tested in the DIII-D tokamak, successfully maintaining low tearing likelihood even under unfavorable conditions like low safety factor and low torque. The controller allowed the plasma to actively track a stable path while maintaining high-performance H-mode operation, which is difficult with traditional preprogrammed control. The AI controller uses total beam power and plasma triangularity as action variables, optimizing for high plasma pressure while keeping tearing likelihood low. The reward function was designed to encourage high plasma pressure under tolerable tearability. The controller was trained using a dynamic model that predicts future plasma pressure and tearing likelihood. The AI controller demonstrated the ability to maintain low tearability and achieve higher time-integrated performance compared to traditional control methods. The study shows that RL can be applied to real-time control of core plasma physics and plasma boundary control. The AI controller was tested under different threshold settings for tearability, with the optimal threshold allowing the plasma to remain stable for longer periods. The controller was also tested under new conditions, including radiofrequency heating, showing its robustness in maintaining low tearability. The results demonstrate the potential of RL for developing stable, high-performance operational scenarios in future tokamaks, particularly for ITER. The study highlights the importance of RL in addressing the challenges of plasma control in fusion energy research.This study presents a deep reinforcement learning (RL) approach to avoid tearing instability in tokamak plasmas, crucial for stable and efficient fusion energy production. Tearing instability, a leading cause of plasma disruptions, is challenging to predict and control, especially in the ITER baseline scenario. The researchers developed a dynamic model to estimate the likelihood of future tearing instability based on diagnostic signals and actuator data. This model was used as a training environment for an RL-based AI controller, enabling automated prevention of tearing instabilities. The AI controller was tested in the DIII-D tokamak, successfully maintaining low tearing likelihood even under unfavorable conditions like low safety factor and low torque. The controller allowed the plasma to actively track a stable path while maintaining high-performance H-mode operation, which is difficult with traditional preprogrammed control. The AI controller uses total beam power and plasma triangularity as action variables, optimizing for high plasma pressure while keeping tearing likelihood low. The reward function was designed to encourage high plasma pressure under tolerable tearability. The controller was trained using a dynamic model that predicts future plasma pressure and tearing likelihood. The AI controller demonstrated the ability to maintain low tearability and achieve higher time-integrated performance compared to traditional control methods. The study shows that RL can be applied to real-time control of core plasma physics and plasma boundary control. The AI controller was tested under different threshold settings for tearability, with the optimal threshold allowing the plasma to remain stable for longer periods. The controller was also tested under new conditions, including radiofrequency heating, showing its robustness in maintaining low tearability. The results demonstrate the potential of RL for developing stable, high-performance operational scenarios in future tokamaks, particularly for ITER. The study highlights the importance of RL in addressing the challenges of plasma control in fusion energy research.