Mechanist: AI as a Scientific Instrument for Discovering the Mechanisms of Intelligence

2026-08-12

AI Summary

This work addresses the growing gap between the rapid advancement of AI systems and our limited understanding of their internal mechanisms and associated risks, which human analysis alone cannot keep pace with. To bridge this gap, the paper introduces Mechanist—the first fully automated, AI-driven framework for mechanistic discovery—treating large language models as scientific instruments to close the loop from behavioral observation and mechanistic explanation to intervention and control. Mechanist integrates a knowledge graph constructed from 13,000 interpretability papers, a corpus of 43 million scientific articles spanning 26 domains, and 32 causal analysis and validation methods. Experiments demonstrate that Mechanist efficiently generates high-value mechanistic hypotheses, reliably conducts automated experiments, and successfully guides the design of DNA sequences with specified functionalities in foundation models, achieving breakthroughs in critical areas such as cross-modal risk propagation and model belief formation.

Abstract

AI models have achieved remarkable success across diverse domains, yet the mechanisms underlying their capabilities and the risks they may pose remain poorly understood. As AI development becomes faster and increasingly automated, mechanistic exploration remains largely manual, widening the gap between what models can do and our ability to understand and control them. To bridge this gap, we introduce Mechanist, an agentic system that uses AI as a scientific instrument for the autonomous discovery of mechanisms underlying AI intelligence. To support autonomous mechanistic discovery, we construct an interpretability-focused knowledge graph of approximately 13,000 papers and integrate it with a multidisciplinary database of 43 million papers spanning 26 fields. We further curate a library of 32 foundational methods for mechanism analysis, causal intervention, and validation. Compared with Claude Code and existing AI-scientist systems, Mechanist generates more valuable mechanism hypotheses and executes experiments more reliably. Mechanist also demonstrates a progression from discovering model behaviors to explaining and controlling AI models. Specifically, Mechanist first uncovers a counterintuitive safety risk in scientific laboratories, showing that unsafe traits can transfer across modalities through apparently safe training data. Mechanist then develops a mechanism theory of belief, revealing how models represent world knowledge, form beliefs, infer the beliefs of others, and how these mechanisms emerge during pretraining. Finally, Mechanist translates these mechanistic insights into practical interventions that improve model performance across diverse scenarios and steer scientific foundation models toward generating DNA sequences with specified properties.

Problem

• mechanistic understanding
• AI interpretability
• intelligence mechanisms
• AI safety risks
• model control

Innovation

• mechanistic interpretability
• AI as scientific instrument
• autonomous discovery
• causal intervention
• foundation models