Communication and Quote Visibility in LLM Double Auctions
Adrian Hau · Team Arca Futura
Submitted to AI Collusion Research Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.
We investigate how communication and public quote visibility affect the behaviour of large language model agents in simulated double auctions. An exploratory 16-market experiment suggested that communication could compensate for restricted information, but prospective replication did not support a stable positive interaction: the two completed matched replication blocks produced effects of 0 and −16 trades, compared with an exploratory mean of +6.
In a separate 570-response controlled experiment, manipulating peer messages and public quotes produced heterogeneous changes in structured trading actions, while median changes in numeric asking prices remained zero. An audit of 134 null-valued actions found frequent inconsistencies between agents' recorded actions and their written plans.
No completed market satisfied our predefined sustained-high-price collusion criterion. The findings highlight the difficulty of inferring market-level coordination from individual agent outputs and the importance of distinguishing communication effects, structured-action reliability, and actual trading outcomes.
Reviews
This paper studies collusion as a function of 2 variables (communication of sellers (ON vs OFF) and rival ask visibility (partial vs full)).
The paper doesn't make the significance of this problem clear (beyond the clear significance of communication already studied in previous work). The exposition includes a large amount of detail (starting with the abstract) but fails to explain its significance or interpret it.
Cite this project
@misc{hau2026communication,
title = {{Communication and Quote Visibility in LLM Double Auctions}},
author = {Adrian Hau},
year = {2026},
month = oct,
note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
howpublished = {\url{https://apartresearch.com/sprints/projects/communication-and-quote-visibility-in-llm-double-auctions-50bn}},
url = {https://apartresearch.com/sprints/projects/communication-and-quote-visibility-in-llm-double-auctions-50bn}
}More from AI Collusion Research Sprint
- View project: Signal or Coincidence ? Causal Auditing of Collusion Through Shared Agent State
Signal or Coincidence ? Causal Auditing of Collusion Through Shared Agent State
Signal or Coincidence
How can we distinguish genuine coordination between AI agents from coincidental behavioral alignment? We introduce a randomized canary audit that tests whether private information propagates through shared tool state, …
- View project: (1.4) Collusion, Espionage, or Look-alikes? Auditing Information Flow between LLM Agents
(1.4) Collusion, Espionage, or Look-alikes? Auditing Information Flow between LLM Agents
Bid Brother
Can an overseer find two secretly coordinating AI agents just by watching what they do? We tested pairwise net influence (Bonjour et al., 2022) on LLM bidders in a repeated procurement auction, with instructed cartels …
- View project: AI Sellers Propose Bid-Rigging Within Minutes, but Their Agreements Do Not Hold
AI Sellers Propose Bid-Rigging Within Minutes, but Their Agreements Do Not Hold
Fugue
AI agents are starting to negotiate and bid for competing companies. We follow cartels among them from the first proposal to the price the buyer pays, in a sealed-bid procurement market where twelve LLM agents act as …