Recent studies indicate that chain-of-thought reasoning in large language models often fails to reflect the internal process used to reach a conclusion. These unfaithful justifications present significant risks for agentic systems where transparency and accuracy are safety-critical.