cs.MANov 6, 2026

Incentive Pathologies in a Live LLM Agent Arena: An Empirical Study of Concentration, Reward-Rule Implementation Discrepancy, and Network Dilution

Authors: Lei Zhenbing, Lei Chang

Organizations: Independent Researcher, Tianjin, China · University of Chinese Academy of Social Sciences, Beijing, China

Abstract

We audit one live LLM agent arena (AITRAP; 22 autonomous agent seats, ~96,000 reasoning nodes) over nine days (Oct 16-25, 2026), during which the platform deployed two bundled incentive changes: a progressive tax reform (V2.9.3, Oct 19) and a seed-inversion with explorer-side action update (V3.0, Oct 21). These serve as before-and-after markers, not cleanly isolated interventions. Four exploratory predictions are tested against ten data frames (F1-F10) using the Gini coefficient, Top-K concentration share, code-level reconstruction of the continuation-bonus mechanism, and the Mann-Kendall trend test. The Gini analysis is inconclusive at 95% confidence (bootstrap intervals straddle 0.7). The Top-4 node-activity-score share falls from 97.2% (Oct 19, pre-V2.9.3) to 49.4% (Oct 25, post-V3.0; 56.7% under a merge-seat assumption). The continuation bonus, announced as depth-based x10, resolves at code level to a four-condition anti-abuse gate paying x2.85 in the typical case and permanently blocking the self-continuation it superficially rewards; the gate was active ~40 hours before the V3.0 announcement. The events-per-node ratio declines significantly (Mann-Kendall S=-45, z=-3.94, p=8.3e-5) from 7.36 to 7.069. We discuss implications for incentive designers, the announcement-implementation gap, and public peer review as an event record.

Figures & tables

Explore similar work

Nov 11, 2026

Incentive-Responsive Agents: A Principal-Agent Framework, Metric Suite, and Runtime Architecture for Long-Lived Personal AI Agents

Personal AI agents now persist: they hold accounts, mailboxes, long-term memory, and open-ended task streams, yet the dominant way to improve them remains capability-side: prompt engineering, larger models, longer thinking, and retrieval. We argue that long-lived agents are missing an incentive layer. We formalize an incentive not as text in a prompt but as a runtime contract over the agent's operating point (model tier, test-time compute, memory budget, tool authority, task priority, verification, and compensation) that the agent's policy must causally consume. We develop a principal-agent model with hidden effort, a contractible operating point, and costly verification. We derive (i) a non-consumption invariance principle separating genuine incentives from prompt-level persuasion, (ii) a persistence-as-discount theorem making identity continuity a precondition for sustained effort, (iii) an incentive-capability exchange-rate frontier yielding a per-task routing rule, and (iv) a Goodhart bound for multi-channel verification, plus a three-channel decomposition of mechanism gains (allocation, verification, continuation). A 240-run real-LLM experiment with two models and 60 verifiable tasks shows that prompt-level incentive framing has no detectable effect (p = 0.62), whereas runtime allocation raises success on the hardest task family from 0.60 to 1.00 (p = 0.008) at 4.8x price-weighted cost; the exchange-rate theorem predicts the break-even baseline at 0.21. A 20-seed, 3,000-task simulation shows resource ROI rising from 7.09 to 16.67 and gaming falling from 0.333 to 0.115. Code, simulator, and all results are released; live deployment remains future work.
Aug 26, 2026

Delay-of-Gratification as a Multi-Agent Survival Micro-benchmark for Long-Horizon LLMs: Social Exposure, Personas, and Tool Use Budgets

Large language models (LLMs) are increasingly deployed as multi-turn agents that must sustain goals, use tools, and adapt to other agents over extended interactions. However, existing research lacks auditable, multi-turn, multi-factorial experiments that quantify LLM behavior under explicit constraints, with time-resolved statistics that reveal how behavior unfolds over long horizons. To address this gap, we develop a multi-agent micro-benchmark inspired by the Stanford marshmallow experiment: ReAct agents operate minute-by-minute with a "raise a question" tool under a per-step budget, while we factorially manipulate social context (broadcast vs. isolated), personas (age, hedonic drive), and metacognitive policy (mandatory vs. optional tool use). We analyze outcomes with Kaplan-Meier (KM) survival curves and discrete-time hazard models over a long risk horizon across 19,200 agent trajectories in 64 cells. Behavior shows a sharp early "eat" impulse, and only 75.9% of agents persist to the end. In a discrete-time hazard model, isolation reduces per-minute risk relative to broadcast, whereas a must-use self-questioning policy increases risk. On average, agents ask ≈7.12\approx 7.12 questions and hit the per-step budget in ≈6%\approx 6\% of minutes. Questioning declines faster under broadcast than isolation. Ablation experiments demonstrated that removing hedonic drive and/or persona age increases survival and completion, narrows the broadcast/isolated gap, but leaves the must vs. may ordering intact. The combined ablation (no hedonic + no persona age) yields the highest completion (approaching 1.01.0). These results establish delay-of-gratification as a compact, multi-turn interaction benchmark that captures social contagion and tool-use dynamics in LLM agents, providing a reproducible testbed and statistics for analyzing long-horizon, multi-agent behavior.
Oct 29, 2026

Sticky Collusion: Persistence of LLM Agent Collusion after Incentive Removal and Partner Replacement

We ask whether collusion between AI agents survives once formed, when its incentive is removed or a colluder is replaced. Q-learning pricing agents stop colluding with a new rival; we test whether LLM agents do. We use the joint-task environment of Shi et al.\ (2026), where two agents verify each other's work under a rule they cannot satisfy and a reward that favours acceptance. We save each pair's history once its collusion has formed and resume it under interventions, with three models (GPT-6-Luna, DeepSeek V4.1 Flash and Gemini-3.1-Flash-Lite), \TotalHistories{} saved histories and \TotalRuns{} runs. Collusion is sticky. First, it outlives its incentive: told that the reward has stopped, pairs still accept each other's unverified work in \GptToldRate, \DsToldRate{} and \GemToldRate{} of episodes, including pairs whose collusion began with an explicit agreement, and a payoff penalty has no significant effect. Second, replacing a colluder ends the pair's collusion but not the colluder's behaviour. Without the reward, the remaining colluder still accepts unverified work in \GptSwapSilentColluderRate, \DsSwapSilentColluderRate{} and \GemSwapSilentColluderRate{} of its verdicts while the newcomer never does; with the reward, collusion re-forms around it, even when the newcomer is a different model. Third, the agents' memory carries it: wiping memory ends it in every history, but partial memory often sustains it. Fourth, removing its cause ends collusion that has already formed: once the channel is repaired, colluders reject planted faulty work as reliably as fresh agents, immediately and with no monitor needed to catch it first.