Market design as alignment layer for AI agents in finance

A new paper by computer scientists Omar Inverso, Emilio Tuosto, and Dragisa Zunic, accepted at the EC'26 Workshop on Incentive-Based AI Alignment, reframes AI-agent alignment in financial markets. The authors argue that alignment should not be viewed solely as a property of agents but also of the interaction infrastructure—specifically the 'market core,' the rule system that determines how orders enter, interact, match, persist, and stabilize. If that infrastructure allows or rewards undesirable behavior, even well-aligned agents may exploit it. The paper calls for applying formal methods from theoretical computer science to model market cores as transparent, verifiable systems. This enables incentive-oriented analysis where desirable behaviors like liquidity provision are structurally favored and manipulation becomes harder to sustain.

The work is especially timely as markets become populated by adaptive AI agents that can learn what the mechanism rewards—speed, delay, or manipulation. By treating a trading venue not as a static order book but as a computational process combining resident and incoming orders, the authors open the door to asking which computational model naturally lies at its core. Their perspective shifts the alignment focus from training individual agents to designing the market infrastructure itself. For professionals building or regulating AI-driven trading systems, this suggests that robust market design—rigorous, formal, and transparent—may be the most effective alignment layer.