Circuit Oracle: Automating Attribution Graph Analysis via Natural-Language Queries

Authors: Hong Kiat Tan*, Shariar Kabir*, Swastik Agrawal, Sai Chereddy, Sriram Balasubramanian   (* equal contribution; ordered alphabetically by first name )

Accepted at the Mechanistic Interpretability (MechInterp) and Computational Learning (CompLearn) workshops at ICML 2026.

Attribution graphs, an emerging tool in mechanistic interpretability, use transcoders to decompose language model computations into sparse interpretable features connected by causal edges. However, turning a graph into a safety-relevant insight requires hours of manual analysis by experts. We introduce Circuit Oracle, a multi-agent system that automates this analysis by autonomously answering natural-language questions about a target model (e.g., “Is this prediction driven by spurious features?”) through multi-hop traversal of the attribution graph. We evaluate Circuit Oracle on three safety-relevant proxy tasks: detecting spurious features in probe circuits, eliciting hidden knowledge from taboo-finetuned models, and jailbreaking via causal interventions. On all three tasks, the oracle is comparable to or exceeds task-specific baselines that do not use the attribution graph. The circuit oracle requires no fine-tuning, as each task is specified by a modular skill (a natural-language prompt) paired with task-specific tools such as transcoder-feature steering, making the framework extensible by construction. Our results suggest that off-the-shelf agents reading attribution graphs through tool calls offer a practical, general-purpose route to automated mechanistic interpretability.

[Paper]

Recommended citation: Hong Kiat Tan, Shariar Kabir, Swastik Agrawal, Sai Chereddy, Sriram Balasubramanian. (2026). "Circuit Oracle: Automating Attribution Graph Analysis via Natural-Language Queries." MechInterp & CompLearn Workshops at ICML 2026.