Catastrophic Forgetting

Catastrophic forgetting is when a neural network loses previously learned knowledge after being trained on new data. For agents, this means fine-tuning or updates can degrade performance on earlier tasks — requiring regression testing and governance review.