Skip to content

Source paper

ARIS: Autonomous Research via Adversarial Multi-Agent Collaboration

Ruofeng Yang, Yongcan Li, Shuai Li
arXiv 2605.03042 · 2026158 upvotes on Hugging Face

Research ideas from this paper (2)

Generated by AI. The paper’s authors did not write these.

AI summary

This paper presents ARIS, an open-source research harness for autonomous machine learning research designed to mitigate the risk of plausible unsupported claims produced by long-horizon agents. ARIS enforces cross-model adversarial collaboration by pairing an executor LLM from one model family with a reviewer LLM from another family across five automated research workflows. The system incorporates an assurance audit cascade, persistent research wiki state, and modular Markdown skills to ensure experimental integrity and manuscript quality.

How it works (AI summary)

ARIS operates across three functional layers: execution, orchestration, and assurance. The execution layer provides over 65 Markdown-defined skills, model integrations via Model Context Protocol (MCP), and a stateful knowledge graph research wiki. The orchestration layer coordinates workflows across discovery, experimentation, writing, and rebuttal phases with adjustable effort levels. The assurance layer enforces experimental integrity through a three-stage audit cascade (integrity checks, result-to-claim mapping, zero-context paper claim audit), scientific prose refinement, proof checking, and visual PDF inspection.