Anterior insula (AI) and dorsal ACC (dACC) are known to process information about pain, loss, adversities, bad, harmful or suboptimal choices and consequences that threaten survival or well-being. Also pregenual ACC (pgACC) is linked to loss and pain, being activated by sad thoughts and regrets. Lateral habenula (LHb) is stimulated by predicted and received pain, discomfort, aversive outcome, loss. Its chronic stimulation makes us feel worse/low and gradually stops us choosing and moving for the suboptimal or punished choices, by direct and indirect (via rostromedial tegmental nucleus, RMTg) inhibition of dorsal raphe nucleus (DRN) and VTA/SNc. The response selectivity of LHb neurons suggests their cortical input from affective and cognitive evaluative regions that make expectations about bad, unpleasant or suboptimal outcomes. Based on these facts I predicted direct dACC, pgACC and AI projections to LHb, which form part of an adversity processing circuit that learns to avoid bad outcomes by suppressing dopamine and serotonin signal. To test this connectivity I used Diffusion Tensor Imaging (DTI). I found dACC, pgACC, AI and caudolateral OFC (clOFC) projections to LHb. I predicted no corticohabenular projections from the reward processing regions: medial OFC (mOFC) and ventral ACC (vACC) because both respond most strongly to good, high valued stimuli and outcomes, inducing dopamine and serotonin release. This lack of LHb projections was confirmed for vACC and likely for mOFC. The surprising findings were the corticohabenular projections from the cognitive prefrontal cortex regions, known for flexible reasoning, planning and combining whatever information are relevant for reaching current goals. I propose that the prefrontohabenular projections provide a teaching signal for value-based choice behavior, to learn to deselect, avoid or inhibit the potentially harmful, low valued or wrong choices, goals, strategies, predictions and ways of doing things, to prevent bad or suboptimal consequences.