mlsanityData to 14 Oct 2026

Prompt-Activation Duality: Improving Activation Steering via Attention-Level Interventions

arXiv 2605.10664 ↗First posted 11 May 2026cs.CL cs.AIcited by 6

On the map: Mechanistic interpretability → Activation steering methods

Abstract (arXiv)

Activation steering controls language model behavior by adding directions to internal representations at inference time, but standard residual-stream steering can fail in stateful dialogue. We identify KV-cache contamination as a key failure mode: steered token states are stored and repeatedly reused, turning a local perturbation into cumulative coherence degradation. To address this challenge, we propose Gated Cropped Attention-Delta steering (GCAD), which extracts steering signals from system-prompt contributions to self-attention and applies them with token-level gating. Across persona-steering experiments, GCAD preserves trait control while substantially improving long-horizon coherence. On the main multi-turn benchmark, GCAD improves average coherence drift from -18.6 to -1.9 and raises turn-10 trait expression from 78.0 to 93.1. These results suggest that activation steering becomes more reliable when interventions follow the prompt-mediated pathways that models already use for behavioral control.

Key facts. Cited by 6 papers in the map (data through 14 Oct 2026). Placed in Activation steering methods, one of 13 directions in Mechanistic interpretability. Builds on 33 of its 35 references that are in the map. First posted on arXiv 11 May 2026.

Builds on

33 of 35 references.

Closest ancestors · 10

Same direction, newest first

  1. 2026Contextual Linear Activation Steering of Language Models27 Apr
  2. 2026What Drives Representation Steering? A Mechanistic Case Study on Steering Refusal9 Apr
  3. 2026Do Personality Traits Interfere? Geometric Limitations of Steering in Large Language Models23 Jan
  4. 2025Understanding (Un)Reliability of Steering Vectors in Language Models28 May
  5. 2024Towards Reliable Evaluation of Behavior Steering Interventions in LLMs22 Oct
  6. 2024Improving Instruction-Following in Language Models through Activation Steering15 Oct
  7. 2024Steering Large Language Models using Conceptors: Improving Addition-Based Activation Engineering9 Oct
  8. 2024Analyzing the Generalization and Reliability of Steering Vectors17 Jul
  9. 2024Extending Activation Steering to Broad Skills and Multiple Behaviours9 Mar
  10. 2023Steering Llama 2 via Contrastive Activation Addition9 Dec

Other references · 23

  1. 2020Language Models are Few-Shot Learnerscited by 32,419
  2. 2022Training language models to follow instructions with human feedbackcited by 15,971
  3. 2024The Llama 3 Herd of Modelscited by 14,715
  4. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,459
  5. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  6. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  7. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,944
  8. 2024Qwen2.5 Technical Reportcited by 4,942
  9. 2021Prefix-Tuning: Optimizing Continuous Prompts for Generationcited by 3,915
  10. 2022FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awarenesscited by 3,673
  11. 2023Efficient Streaming Language Models with Attention Sinkscited by 2,186
  12. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,336
11 more
  1. 2023Inference-Time Intervention: Eliciting Truthful Answers from a Language Modelcited by 1,302
  2. 2023Instruction-Following Evaluation for Large Language Modelscited by 1,228
  3. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 1,032
  4. 2024The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructionscited by 493
  5. 2023Scissorhands: Exploiting the Persistence of Importance Hypothesis for LLM KV Cache Compression at Test Timecited by 439
  6. 2025Persona Vectors: Monitoring and Controlling Character Traits in Language Modelscited by 345
  7. 2023In-context Vectors: Making In Context Learning More Effective and Controllable Through Latent Space Steeringcited by 245
  8. 2025MultiChallenge: A Realistic Multi-Turn Conversation Evaluation Benchmark Challenging to Frontier LLMscited by 180
  9. 2025Taxonomy, Opportunities, and Challenges of Representation Engineering for Large Language Modelscited by 34
  10. 2025Representation Engineering for Large-Language Models: Survey and Research Challengescited by 21
  11. 2025Steering Multimodal Large Language Models Decoding for Context-Aware Safetycited by 7

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 6.

6 papers.

  1. 2026Activation Steering Transfer to Agents: One Gain Ratio Does Not Identify Potency and Efficacycited by 0
  2. 2026Gathered, Not Admitted: How Attention Brings a Latent Variable into Verbalizable Formcited by 0
  3. 2026K/V-Cache Interventions Dissociate Representation Alignment from Persona Expression in Decoder-Only Language Modelscited by 0
  4. 2026One Token Can Be Enough: Bridging Prompting and Activation Steering with Prefix Steeringcited by 0
  5. 2026Loud and Clear: Dynamic Activation Steering for Improving Speech Intelligibility in Noisy Environmentscited by 0
  6. 2026Contrastive Covariance Subspaces for Activation Steeringcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Steer Like the LLM: Activation Steering that Mimics Prompting
  2. 2026Memory Inception: Latent-Space KV Cache Manipulation for Steering LLMs
  3. 2026Where Steering Signals Come From: Activation Source Selection in Activation Steering
  4. 2026Activation Steering via Generative Causal Mediation
  5. 2026Activation Steering Induces Emergent Misalignment: A More Comprehensive Evaluation
  6. 2026Steer2Edit: From Activation Steering to Component-Level Editing
  7. 2026CBMAS: Cognitive Behavioral Modeling via Activation Steering
  8. 2026Beyond Steering Vector: Flow-based Activation Steering for Inference-Time Intervention