Abstract: Epilepsy is the second most common neurological disorder after stroke, and affects at least 50 million people world-wide. Nearly 40% of new onset epilepsy cases are pharmaco-resistant, with seizures that cannot be fully controlled. Thus, there are nearly 25 million people world-wide that could benefit from alternative epilepsy treatment paradigms. The goal of this study was to close the loop between seizure prediction and seizure control via timely deep brain stimulation in a rat model of chronic epilepsy. First, seizure predictability was evaluated by nonlinear dynamical analysis of EEG recorded from chronically epileptic rats, in terms of its sensitivity and specificity. Early detection of preictal dynamical entrainment, as revealed by convergence of measures of chaos at critical brain sites was the key for this specific aim. Results showed predictability of seizures with mean sensitivity of 74.6%, specificity of 0.164 false predictions per hour, and a mean prediction time of 63.7 minutes prior to seizure onset. Second, the effect of deep brain stimulation on epileptic brain dynamics was evaluated over a range of frequency and intensity of stimulation. Results showed that high frequency stimulation (≥130 Hz) at maximally tolerated current levels could disentrain the epileptic brain dynamics for up to 30 minutes following the stimulus. Lastly, automated seizure prediction was combined with a control system that, in real-time provided deep brain stimulation to the thalamus to avert the occurrence of seizures long before their expected onset. Significant reduction of seizure frequency (>50%) with this automated control system was evident in 33% of the rats, which also showed improvement over the efficacy of a periodic control paradigm in the same animals with identical stimulation parameters. Successful automated control of seizures was highly correlated with disentrainment in global brain dynamics. When seizure control was not achieved, there was no sign of desynchronization in global brain dynamics. These findings show that (a) seizure control using the proposed scheme appears to be feasible, and (b) there is room for improvement, including a more accurate, adaptive identification of the critical brain sites responsible for seizure generation.