Do Chess Explanations Reflect Model Decisions? Behavioral and Token-Level Tests of LLM Reasoning Faithfulness
Factual evidence
What the source reports
Research across 200 chess puzzles shows LLM explanations often fail to faithfully reflect the model's actual token-level move decisions.
Inspect the evidence
- Inclusion basis
- Enterprise AI
- Publisher and source type
- arXiv cs.CL — Computation and Language · RESEARCH
- Published by source
- 22 September 2026
- Collected by OneBench
- 23 Sept 2026, 03:01 UK
Stored source excerpt
arXiv:2609.22245v1 Announce Type: new Abstract: Large language models can produce fluent explanations for chess moves, but plausible language does not necessarily reflect the reasoning behind…
Short excerpt from the collected text, not the full source. Use the source link to read it in context.
The factual summary is a OneBench synthesis, not a quotation or independent verification. Collection time is not publication time. Open the source for its full context; related reporting can share the same underlying announcement.
OneBench interpretation
Institutional assessment
So what
Demonstrating that LLM-generated rationale can diverge from underlying decision mechanics challenges relying on generated text for regulatory explainability.
Do what
Review explainability testing methodologies with the team responsible for model risk management when validating LLMs for decision support.