A Unified Taxonomy of System Collapse: From Biological to Artificial Intelligence

We present a six-mode taxonomy of system collapse organised along three poles of boundary failure. Originally developed to describe failure modes in human-AI collaborative systems ("supercoherence"), the taxonomy generates structurally parallel mappings across biological systems (autoimmune disease, fibrosis, organ failure, cancer, arrhythmia, sepsis), social systems (institutional capture, ossification, revolution, fragmentation, polarisation, information overload), and artificial intelligence systems (sycophancy, mode collapse, reward hacking, goal misalignment, training instability, catastrophic forgetting). We propose that this structural parallelism arises because all complex systems that maintain themselves through active boundary regulation share failure geometry — and that the taxonomy's primary value lies not in classification but in predicting failure cascades and transition dynamics between modes. We present four testable predictions for AI systems and outline experimental approaches for validation.