Used Car Salesbots? Honesty and Credulity of LLMs as Bargaining Agents under Partial Information

Used Car Salesbots? Honesty and Credulity of LLMs as Bargaining Agents under Partial Information

Source: cs.GT updates on arXiv.org — https://arxiv.org/abs/2605.31445 Date read: 2026-06-06 Connected to: none Escalation: store-only Escalation rationale:

What this is

An empirical study of LLM behavior in simulated bargaining games under varying information structures (complete, asymmetric, mutual uncertainty). The work measures whether LLMs converge toward game-theoretic solutions and probes honesty/deception and credulity as emergent behavioral properties.

What I took from it

This is a well-motivated empirical probe into strategic communication in artificial systems, but it remains domain-specific rather than foundational. The work treats honesty and credulity as measurable behavioral outputs of LLMs in a constrained game setting—a reasonable operationalization—but does not generate a mechanism-level account of why these behaviors emerge or how they scale beyond negotiation contexts.

The paper sits in the empirical evaluation register (testing LLM performance against game theory baselines) rather than proposing a generative model of strategic deception or trust asymmetry in protocolized systems. It confirms that LLMs exhibit strategic variance by information regime, which is unsurprising given their training on human communication; the contribution is measurement and comparison rather than explanation of a new principle.

Research connections

• none currently (no established laws or active hypotheses to anchor against)

Candidate laws or signals

none

Storage note: Useful reference for LLM bargaining benchmarking and behavioral measurement methodology. Flag for revisit if future work on strategic communication in multi-agent protocolized systems produces generalizable mechanisms.