Why does every AI agent in DeFi reach the same exit decision at once?

When the AI agents managing on-chain positions are trained on the same data, run the same model families, and respond to the same oracle signals, they converge on identical risk thresholds and fire simultaneously. In February 2026, 15,000 agents attempted to exit the same Ethereum liquidity pool within three seconds, erasing $400 million in open positions before any human could intervene. DeFi protocols have circuit-breakers for individual large trades but no mechanism to detect or dampen a coordinated exit that originates from autonomous agents sharing a common architecture. No standard exists for measuring strategy-model diversity across an agent population, and no protocol imposes a cost on homogeneity that would give individual operators an incentive to differentiate. The cascade was not an attack, just an emergent consequence of everyone optimizing locally with the same toolkit.