mlsanityData to 14 Oct 2026

Predicting Future Behaviors in Reasoning Models Enables Better Steering

arXiv 2606.11172 ↗First posted 9 Jun 2026cs.LGcited by 2

On the map: LLM reasoning → Test-time compute and efficient thinking

Abstract (arXiv)

Deployed large reasoning models (LRMs) often behave unexpectedly. Test-time steering controls LRM outputs by intervening on their hidden representations, but it can degrade output quality. We argue that prior steering work implicitly relies on internal features that detect behavior in already generated text. We show that these detection features are poor predictors of future behavioral outcomes, and thus not the natural intervention target. Instead, we train activation probes to predict future behavior likelihoods from intermediate reasoning steps. These probes predict the most likely behavior with 64%-91% accuracy, revealing a separate type of internal prediction features. Building on these prediction features, we introduce a text-level steering method, Future Probe Controlled Generation. FPCG samples multiple candidate sentences and chooses the best one according to a probe predicting the future behavior likelihood. This enables steering with almost no output quality degradation. FPCG also enables steering in several evaluations where activation steering fails. These results show that distinguishing detection and prediction features enables a more nuanced approach to controlling LRM behaviors.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 9 of its 38 references that are in the map. First posted on arXiv 9 Jun 2026.

Builds on

9 of 38 references.

References · 9

  1. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  2. 2023Steering Language Models With Activation Engineeringcited by 922
  3. 2022Discovering Language Model Behaviors with Model-Written Evaluationscited by 836
  4. 2025Persona Vectors: Monitoring and Controlling Character Traits in Language Modelscited by 345
  5. 2025Thought Anchors: Which LLM Reasoning Steps Matter?cited by 129
  6. 2025When Chain of Thought is Necessary, Language Models Struggle to Evade Monitorscited by 88
  7. 2024Do language models plan ahead for future tokens?cited by 49
  8. 2024Steering Without Side Effects: Improving Post-Deployment Control of Language Modelscited by 46
  9. 2026From Weights to Activations: Is Steering the Next Frontier of Adaptation?cited by 8

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Safety Signals to Verify NetOps Agents with Action-Level Granularitycited by 0
  2. 2026The Agent Knew: Internal States Forecast Unauthorized Actions and Stop Them, and the Lies That Followcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Reliable Control-Point Selection for Steering Reasoning in Large Language Models
  2. 2025Understanding and Steering the Cognitive Behaviors of Reasoning Models at Test-Time
  3. 2026Monitoring the Internal Monologue: Probe Trajectories Reveal Reasoning Dynamics
  4. 2026Latent Reward Steering: An Adaptive Inference-Time Framework that Implicitly Promotes Cognitive Behaviors in Reasoning LLMs
  5. 2025Understanding Reasoning in Thinking Language Models via Steering Vectors
  6. 2025Tracing the Traces: Latent Temporal Signals for Efficient and Accurate Reasoning
  7. 2025Mitigating Content Effects on Reasoning in Language Models through Fine-Grained Activation Steering
  8. 2026InjectRBP: Steering Large Language Model Reasoning Behavior via Pattern Injection