Reasoning-Token Spikes Under Prompted Untruthful Responding in Large Language Models
The paper tests whether reasoning-capable LLMs use more hidden “thinking” tokens when prompted to answer falsely or truth-indifferently, finding across three models and 210 multiple-choice questions that untruthful response policies reliably increase reasoning-token counts versus truth-directed answering. The neat idea is that token-count spikes are a cheap, content-independent signal that might still work when chain-of-thought text is hidden or untrustworthy, but the result is only a group-level proof of concept for explicitly prompted lying, not evidence yet for reliable instance-level deception detection, spontaneous misalignment, or robustness against adversarial models.
Abstract
Monitoring the chain-of-thought of reasoning artificial intelligence (AI) models remains a key approach to detecting deception and other forms of misbehavior in such models. However, semantic chain-of-thought monitoring depends on reasoning traces being legible and sufficiently faithful to the underlying computations that produced the model's behavior, not to mention accessible. Moreover, there is increasing evidence that chain-of-thought outputs may soon become illegible or unfaithful, if they even remain accessible. Based on cognitive load theory, we investigate a lower-bandwidth signal -- the number of reasoning tokens generated -- which does not require access to the content of the reasoning trace. Three reasoning-capable large language models answered 210 multiple-choice questions -- across analytic, descriptive, and normative reasoning types as well as moral and non-moral domains -- under system prompts instructing them to respond truthfully, falsely, or without regard for truth. Across all three models, truth-directed responding elicited fewer reasoning tokens than both lie-directed and truth-indifferent responding. These findings show that explicitly prompted untruthful response policies can produce robust group-level differences in test-time reasoning-token use. While not yet establishing reasoning-token count as a detector of spontaneous deception or general misalignment, our results are a proof of concept that it can serve as a simple, content-independent candidate signal for differentiating untruthful from truthful model behavior when raw reasoning traces are unavailable or unreliable. Future work should test instance-level detection rates, out-of-distribution generalization, learned deceptive policies, hidden objectives, and robustness under adversarial pressure.