Causal Chain of Thought: Mitigating Attention Leakage via Context Isolation

Gabriele Dominici, Teague McMillan, Fazl Barez, Marc Langheinrich, Martin Gjoreski|October 22, 2026arXiv

Key Takeaway

Chain-of-thought explanations can be misleading because models can look up answers directly from context while generating plausible justifications. Isolating reasoning steps to only their causal dependencies makes explanations more faithful and reasoning more transparent.

Summary

This paper introduces CausalCoT, a method that improves chain-of-thought reasoning by breaking problems into isolated sub-questions that only see their direct dependencies.

reasoning evaluation safety

Key Terms

chain-of-thought attention-leakage directed-acyclic-graph causal-sensitivity context-isolation