ARIS: Open-Source Research Harness for Autonomous ML Workflows
ARIS (Auto-Research-in-sleep) is an open-source research harness designed to coordinate machine-learning research workflows through cross-model adversarial collaboration. The system, described in arXiv:2605.03042, uses an executor model to drive progress while a reviewer from a different model family critiques intermediate outputs to prevent plausible unsupported success—a failure mode where long-running agents produce claims with incomplete or misreported evidential support. The architecture emphasizes assurance mechanisms and early deployment experience, highlighting that agent performance depends on both model weights and the harness governing information storage, retrieval, and presentation.
Key facts
- ARIS is an open-source research harness for autonomous research.
- It uses cross-model adversarial collaboration as a default configuration.
- An executor model drives progress while a reviewer from a different model family critiques outputs.
- The central failure mode addressed is plausible unsupported success.
- The system is described in arXiv:2605.03042.
- Agent performance depends on model weights and the harness.
- The harness governs information storage, retrieval, and presentation.
- Early deployment experience is included in the report.
Entities
Institutions
- arXiv