Large language models are increasingly used as interacting agents, but it remains unclear how robust their coordination is when public communication is unreliable. We study this question in iterated N-player Stag Hunt games played by homogeneous LLM groups under controlled programmatic action inversion, which changes both the public transcript and the actions used for execution. Across an experimental grid spanning group sizes, coordination thresholds, corruption levels, and seven LLMs, we observe three main patterns. First, honest agents’ pre-flip Stag choices decline as corruption increases, but the sharp fall in public success is primarily mechanical. In the focal N=5,M=3 setting, pre-flip success remains 78% at 80% corruption, while public success falls to 12%. Second, honest choices are associated with the public history available at decision time, particularly under high corruption. Third, three threshold-style public-report benchmarks yield similar action-match rates to the LLM agents, showing substantial descriptive agreement between LLM decisions and these benchmarks. Overall, our results show that original choices, public actions, and executed outcomes must be separated when evaluating multi-agent robustness, as corrupted communication can severely and predictably degrade mutually beneficial cooperation.

Online advertising governance faces significant challenges due to the non-stationary nature of regulatory policies, where emerging mandates (e.g., restrictions on education or aesthetic anxiety) create severe label inconsistencies and reasoning ambiguities in historical datasets. In this paper, we propose ARGUS, a policy-adaptive governance system that enables evolving reinforcement through multi-agent adversarial umpiring. ARGUS addresses the sparsity of new policy data by employing a three-stage framework: (1) Policy Seeding for initial perception; (2) Adversarial Label Rectification, which utilizes a ”Prosecutor-Defender-Umpire” architecture to resolve conflicts between stale labels and new mandates; and (3) Latent Knowledge Discovery, which employs a tripartite dialectical discussion to unearth sophisticated, “gray-area” violations. By leveraging RAG-enhanced policy knowledge and Chain-of-Thought synthesis as dynamic rewards for reinforcement learning, ARGUS synchronizes its reasoning pathways with evolving regulations. Extensive experiments on both industrial and public datasets demonstrate that ARGUS significantly outperforms traditional fine-tuning baselines, achieving superior policy-adaptive learning with minimal gold data.

In this system description, we describe our process and the systems that we created for the subtasks A monolingual, A multilingual, and B forthe SemEval-2024 Task 8: Multigenerator, Multidomain, and Multilingual Black-Box MachineGenerated Text Detection. This shared task aimsat detecting and differentiating between machinegenerated text and human-written text. SubtaskA is focused on detecting if a text is machinegenerated or human-written both in a monolingualand a multilingual setting. Subtask B is also focused on detecting if a text is human-written ormachine-generated, though it takes it one step further by also requiring the detection of the correct language model used for generating the text.For the monolingual aspects of this task, our approach is centered around fine-tuning a debertav3-large LM. For the multilingual setting, we created an ensemble model utilizing different monolingual models and a language identification toolto classify each text. We also experiment with thegeneration of extra training data. Our results showthat the generation of extra data aids our modelsand leads to an increase in accuracy.