Behavior Cue Reasoning: Monitorable Reasoning Improves Efficiency and Safety through Oversight

TL;DR

Behavior Cue Reasoning introduces special token signals in LLMs to enhance oversight, enabling more controllable reasoning and significantly reducing unsafe actions without sacrificing performance.

Contribution

We propose Behavior Cues, a novel training method that makes LLM reasoning more monitorable and controllable, improving safety and efficiency during complex reasoning tasks.

Findings

Behavior Cues enable up to 50% reduction in reasoning tokens.

They recover safe actions in 80% of unsafe reasoning traces.

No performance cost observed across multiple models and domains.

Abstract

Reasoning in Large Language Models (LLMs) poses a challenge for oversight as many misaligned behaviors do not surface until reasoning concludes. To address this, we introduce Behavior Cue Reasoning for making LLM reasoning more controllable and monitorable. Behavior Cues are special token sequences that a model is trained to emit immediately before specific implicit and explicit behaviors, acting as dual purpose signal and control levers. When fine-tuning a weaker external monitor with Reinforcement Learning for reasoning oversight, a compressed view of only information surfaced by Behavior Cues is sufficient signal for the monitor to prune up to 50% of otherwise wasted reasoning tokens in complex math problem solving. When leveraged by an almost optimal rule-based monitor in an environment where excessive constraint violations results in failure, Behavior Cues allows for the recovery of safe actions from 80% of reasoning traces that would otherwise end with the proposal of an unsafe action, more than doubling the success rate from 46% to 96%. Through evaluation across two model families and three domains, we show that Behavior Cue Reasoning improves reasoning monitorability and controllability with no cost to performance. More broadly, our work progresses scalable oversight by demonstrating how the monitored model itself can be trained to reason more tractably to oversight. Code: https://github.com/christopherzc/behavior-cues