mlsanityData to 14 Oct 2026

When is Your LLM Steerable?

arXiv 2606.11599 ↗First posted 10 Jun 2026cs.CL cs.LGcited by 3

On the map: Mechanistic interpretability → Activation steering methods

Abstract (arXiv)

Activation steering offers a lightweight approach to control language models' behavior at inference time, but whether it succeeds or fails heavily depends on the prompt, concept, model, and steering configuration. Finding the regime and boundaries of successful steering typically requires expensive grid searches and post-hoc evaluation of full autoregressive rollouts. In this work, we investigate whether steerability can be predicted from the model's internal states at the beginning of the generation process, e.g., after generating the first few tokens, and how to leverage such a predictor to improve steering success rate. To this end, we first introduce ASTEER, a testbed including 1.4M steered generations, spanning 150 concepts with each steering success/failure labeled. Leveraging this testbed, we analyze the model's early decoding dynamics by extracting features that compare hidden states before and after steering across layers and initial decoding steps. These features help us understand how steering's effects propagate along layers and token positions, which provide key information for steerability prediction. We then train a Gradient Boosting Decision Trees (GBDT) classifier on these features to predict whether an intervention will under-steer, succeed, or over-steer without requiring full rollout. Our predictor achieves around 0.7 macro-F1 score on unseen concepts, demonstrating that early hidden states encode substantial, structured information about eventual steering efficacy. We further leverage this steerability predictor as guidance for steering strength searching, achieving near-optimal performance with a small fraction of decoding cost.

Key facts. Cited by 3 papers in the map (data through 14 Oct 2026). Placed in Activation steering methods, one of 13 directions in Mechanistic interpretability. Builds on 23 of its 59 references that are in the map. First posted on arXiv 10 Jun 2026.

Builds on

23 of 59 references.

Closest ancestors · 4

Same direction, newest first

  1. 2023Steering Llama 2 via Contrastive Activation Addition9 Dec
  2. 2023Improving Activation Steering in Language Models with Mean-Centring6 Dec
  3. 2023Steering Language Models With Activation Engineering20 Aug
  4. 2022Extracting Latent Steering Vectors from Pretrained Language Models10 May

Other references · 19

  1. 2016XGBoost: A Scalable Tree Boosting Systemcited by 5,166
  2. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,336
  3. 2023Inference-Time Intervention: Eliciting Truthful Answers from a Language Modelcited by 1,302
  4. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 1,032
  5. 2025AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoderscited by 218
  6. 2024Programming Refusal with Conditional Activation Steeringcited by 206
  7. 2025Reasoning Models Know When They're Right: Probing Hidden States for Self-Verificationcited by 159
  8. 2024LLM Internal States Reveal Hallucination Risk Faced With a Querycited by 82
  9. 2025AdaSteer: Your Aligned LLM is Inherently an Adaptive Jailbreak Defendercited by 38
  10. 2024Activation Scaling for Steering and Interpreting Language Modelscited by 35
  11. 2025To Steer or Not to Steer? Mechanistic Error Reduction with Abstention for Language Modelscited by 29
  12. 2024Beyond Single Concept Vector: Modeling Concept Subspace in LLMs with Gaussian Distributioncited by 27
7 more
  1. 2025Activation Steering for Chain-of-Thought Compressioncited by 24
  2. 2025Can We Predict Alignment Before Models Finish Thinking? Towards Monitoring Misaligned Reasoning Modelscited by 17
  3. 2024Investigating Bias Representations in Llama 2 Chat via Activation Steeringcited by 14
  4. 2024FactCheckmate: Preemptively Detecting and Mitigating Hallucinations in LMscited by 11
  5. 2024Enhancing Multiple Dimensions of Trustworthiness in LLMs via Sparse Activation Controlcited by 5
  6. 2025Bleeding Pathways: Vanishing Discriminability in LLM Hidden States Fuels Jailbreak Attackscited by 5
  7. 2025Stop Before You Fail: Operational Capability Boundaries for Mitigating Unproductive Reasoning in Large Reasoning Modelscited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 3.

3 papers.

  1. 2026Activation Steering Transfer to Agents: One Gain Ratio Does Not Identify Potency and Efficacycited by 0
  2. 2026MetaSteer: Context-Conditioned, nonlinear Steering via Attention-Projection Adaptationcited by 0
  3. 2026On the Steering Dimensionality of Refusal in Language Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Mechanistic Indicators of Steering Effectiveness in Large Language Models
  2. 2026Does Steering Break Your Model? A Multi-Dimensional Evaluation Suite for LLM Steering Methods
  3. 2026Genesis Steering: Better Activation Steering Through Training Data
  4. 2025Understanding (Un)Reliability of Steering Vectors in Language Models
  5. 2025Steering When Necessary: Flexible Steering Large Language Models with Backtracking
  6. 2025Interpretable Steering of Large Language Models with Feature Guided Activation Additions
  7. 2026COLD-Steer: Steering Large Language Models via In-Context One-step Learning Dynamics
  8. 2026Fine-Grained Activation Steering: Steering Less, Achieving More