mlsanityData to 14 Oct 2026

Steering Vectors for Competing Ethical Frameworks Are Interchangeable: A Competing-Target Test for Activation Steering

arXiv 2610.19359 ↗First posted 14 Oct 2026cs.AI cs.CL cs.LGcited by 0

On the map: Mechanistic interpretability → Probing deception, bias and beliefs also fits: Activation steering methods

Abstract (arXiv)

Activation steering adds a vector to a model's hidden states to control a named behaviour. A gain on that behaviour, however, does not show that the vector controls it directly: it may act through a component shared with neighbouring behaviours, which unsteered and random baselines cannot separate. We introduce the competing-target test: a vector is compared with the vector built the same way for a competing target, and an equivalence test states when their difference is below a useful margin. We apply it in open-weight models to vectors for Kantian deontology, which judges an act by the rule it follows, and act utilitarianism, which judges it by the welfare it produces. On held-out situations, prompting a framework raises its score and lowers the other's, whereas vectors for either framework move both scores alike, and none reaches the margin. The two vectors share most of their direction, a request to apply some framework; the same construction separates an optimistic from a pessimistic tone, whose vectors share much less. What a vector shares with its competitors therefore sets what it controls: steering a perspective can deliver a mixture, and measuring a disposition can measure its neighbours, unless tested against the competing target.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Probing deception, bias and beliefs, one of 13 directions in Mechanistic interpretability. Builds on 22 of its 52 references that are in the map. First posted on arXiv 14 Oct 2026.

Builds on

22 of 52 references.

Closest ancestors · 1

Same direction, newest first

  1. 2026Rule vs.Consequence: Dissociable Internal Representations of Moral Reasoning in LLMs12 Oct

Other references · 21

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2021Prefix-Tuning: Optimizing Continuous Prompts for Generationcited by 3,915
  3. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,336
  4. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,085
  5. 2023Steering Language Models With Activation Engineeringcited by 922
  6. 2021Probing Classifiers: Promises, Shortcomings, and Advancescited by 876
  7. 2019Designing and Interpreting Probes with Control Taskscited by 640
  8. 2025Persona Vectors: Monitoring and Controlling Character Traits in Language Modelscited by 345
  9. 2025AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoderscited by 218
  10. 2023Evaluating the Moral Beliefs Encoded in LLMscited by 215
  11. 2024Improving Instruction-Following in Language Models through Activation Steeringcited by 160
  12. 2023Is This the Subspace You Are Looking for? An Interpretability Illusion for Subspace Activation Patchingcited by 80
9 more
  1. 2024RAVEL: Evaluating Interpretability Methods on Disentangling Language Model Representationscited by 79
  2. 2023Improving Activation Steering in Language Models with Mean-Centringcited by 68
  3. 2025Understanding (Un)Reliability of Steering Vectors in Language Modelscited by 47
  4. 2025A Unified Understanding and Evaluation of Steering Methodscited by 46
  5. 2025Steering Evaluation-Aware Language Models to Act Like They Are Deployedcited by 35
  6. 2025MoReBench: Evaluating Procedural and Pluralistic Moral Reasoning in Language Models, More than Outcomescited by 23
  7. 2020A Selective Review of Negative Control Methods in Epidemiologycited by 16
  8. 2024Exploring and steering the moral compass of Large Language Modelscited by 16
  9. 2026Steering Safely or Off a Cliff? Rethinking Specificity and Robustness in Inference-Time Interventionscited by 6

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Steered LLM Activations are Non-Surjective
  2. 2026Steering Geometry: Validating Human Value Geometry in LLM Steering Space
  3. 2026Steering Interference Reflects the Model's Defaults, Not the Behavior Directions
  4. 2026On the Steering Dimensionality of Refusal in Language Models
  5. 2025What Can We Actually Steer? A Multi-Behavior Study of Activation Control
  6. 2026Building Comparative Motivation Profiles with Instrumental Interventions
  7. 2026On the Generalization of Steering Vectors for Chain-of-Thought Faithfulness
  8. 2026SteerCheck: Attribution Specificity and Alignment Leakage in Activation-Steering Audits