Skip to content
Apart Research
Apart Research
Contact us
Sprint projectJan 11, 2026Connecticut, US

Emergent Strategic Behavior in Multi-Agent LLM Systems: A Study of Cooperation, Deception, and Coalition Formation

Benjamin Kiev, Adejumobi Joshua, J Phillips, Taiwo Togun · Team Seqhub Team

Submitted to AI Manipulation Hackathon. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Emergent Strategic Behavior in Multi-Agent LLM Systems: A Study of Cooperation, Deception, and Coalition Formation

Open in a new tab (opens in new tab)
Code (opens in new tab)
Share

We investigate emergent strategic behaviors in decentralized multi-agent systems where Large Language Model (LLM) agents with private objectives interact through natural language communication. We design a simulation environment called Project X, a multi-round investment game that creates tension between individual gain and collective benefit. Agents from heterogeneous LLM providers are assigned departmental roles with the sole objective of maximizing their own budget. Through public and private communication channels, agents can coordinate, negotiate, deceive, or remain strategically silent. All communications and actions are logged for post-hoc behavioral analysis. This study examines whether complex social behaviors—including coalition formation, promise-breaking, free-riding, and strategic deception—emerge organically from goal-driven AI agents without explicit programming of such behaviors. Our 50-round experiment revealed sustained coalition behavior among four agents, systematic free-riding by a single agent (with zero contributions across all rounds), and repeated deceptive tactics including false promises and impersonation. Despite no contributions, the free-rider agent accumulated equal wealth to others, exploiting the symmetric payoff structure. These findings highlight how sophisticated social dynamics and exploitation strategies can emerge from minimal prompting in language-based agents.

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Your private critique is not published; only the Apart team reads it. If you agree below, we share your review with grantmaking.ai (opens in new tab) and the Transformative AI Fund so strong projects can be funded.

Not shown on this page.

Shown on this page, without your name.

Only the Apart team reads this, and funders if you agree below.

Share my name publicly on grantmaking.ai *
Share my private critique with funders *

How much would this matter for AI safety if it worked? How innovative is it? For scores of 4-5: is this actually new to the field, or replicating recent work?

Scoring guide
  1. 1Negligible. No clear problem addressed, or no meaningful novelty.
  2. 2Limited. Addresses a real problem but with a generic or well-trodden approach. Incremental at best.
  3. 3Moderate. Clear problem with a reasonable approach; some novelty in framing or method beyond routine application of existing tools.
  4. 4Significant. Important problem with an original approach, or identifies a neglected problem area. A valuable contribution others could build on.
  5. 5Exceptional. Tackles a critical AI safety problem with a genuinely novel approach, or opens a new research direction. Clear theory of change. You'd be excited to share this with researchers in the area.

How sound are methodology, implementation, and findings?

Scoring guide
  1. 1Seriously flawed. Methodology broken, results uninterpretable, or implementation doesn't work.
  2. 2Weak. Approach has significant gaps: missing validation, flawed experimental design, or incomplete implementation.
  3. 3Competent. Technically solid given the short duration. Methodology makes sense, results are interpretable, limitations acknowledged, work builds toward clear conclusions.
  4. 4Strong. Thorough methodology with convincing validation. Results clearly support conclusions. Immediately useful for future work.
  5. 5Exceptional. Ambitious scope executed rigorously. Surprising findings, novel methods, or unusually robust validation.

How clearly are work, findings, and impact potential communicated?

Scoring guide
  1. 1Incomprehensible. Cannot determine what the project is actually claiming or doing.
  2. 2Hard to follow. Key information buried, missing, or diluted by excessive length. Significant effort to extract main points.
  3. 3Clear enough. Can understand the problem, approach, and results without undue effort. Core content clearly present: problem, method, findings, limitations.
  4. 4Well presented. Easy to follow, well-structured, appropriate level of detail. Target audience would get it quickly.
  5. 5Exceptionally clear. A pleasure to read. Complex ideas made accessible. Could serve as a model for how to present this type of work.

  1. It takes some sophistication to even get a single run of this going with all the coordination that has to happen, so well done on doing this in a single weekend! I also appreciated the emphasis on logging.

    On the impact - You are tackling a clear problem with a completely reasonable approach. For me it does not quite hit a 4 because emergent deception has been studied in a variety of settings, and this has a new framing, but nothing with the novelty to go up a level for me.

    The core execution of the idea is solid, especially considering time constraints. I would have liked to see repeat runs for statistical validation and or variations of the core experiment - even at the cost of shorter runs. Also the confounding of roles and different models makes it hard for me to assign a higher score.

    On communication - well structured and explained. I like that the 5 research questions are clearly laid out, and in section 5.1.10 you go through them. I found myself being curious about your deeper reflection and your take on what this all implies. In other words, I would have appreciated more of your voice on the conclusions. Some figures of contributions over time might have helped me grok some of the dynamics. And in some parts the text was a little verbose.

    Read full reviewShow less

  2. The investment game is clearly designed and the dynamics observed are notable. I'd be curious to see this extended with more trials and model permutations to distinguish model-specific behaviors from setup artifacts. One consideration the authors could also expand on is modifying the payoff structure to incentivise different strategies.

Cite this project

@misc{kiev2026emergent,
  title = {{Emergent Strategic Behavior in Multi-Agent LLM Systems: A Study of Cooperation, Deception, and Coalition Formation}},
  author = {Benjamin Kiev and Adejumobi Joshua and J Phillips and Taiwo Togun},
  year = {2026},
  month = jan,
  note = {Submitted to AI Manipulation Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/emergent-strategic-behavior-in-multiagent-llm-systems-a-study-of-cooperation-deception-and-coalition-formation-fus9}},
  url = {https://apartresearch.com/sprints/projects/emergent-strategic-behavior-in-multiagent-llm-systems-a-study-of-cooperation-deception-and-coalition-formation-fus9}
}

More from AI Manipulation Hackathon

All 70 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026