Counterparty Modeling is Not Strategy: The Limits of LLM Negotiators

In three linesResearchers test LLM agents' negotiation abilities in a controlled multi-attribute bargaining environment. Agents accurately model counterparty preferences but fail to convert this knowledge into winning strategy. Final agreements are driven by opening anchors rather than actual utility weights.

Summary generated by Claude — human-verified