Overview

Activation steering studies how edits to a model’s hidden activations can control its behavior at inference time, enabling changes without retraining the full model. Common approaches add or remove contrastive steering vectors, target sparse autoencoder features, neurons or attention heads, and use conditional or feedback-based controllers to adapt interventions to the input or generation state. Researchers apply these methods to language tasks such as reasoning, refusal, factuality and hallucination mitigation, as well as image and video generation, speech, and embodied agents; evaluations include benchmarks such as GSM8K, TruthfulQA, POPE and LIBERO. A central challenge is that readable or decodable features are not necessarily controllable, and effective edits can vary by context and architecture or cause collateral changes to fluency, utility and safety.

Latest papers

417 papers59 papers in the last four weeks, down 9% on the four weeks before. 0.4% of all new papers.

Oct 14, 2026

MAPLE: A Multi-Axis Parametric Knowledge Extraction Lens for LLM Evaluation

LLMs absorb vast knowledge from training data, yet their inherently probabilistic nature offers no guarantee that this knowledge will surface in their outputs. To bridge this gap, Knowledge Extraction Enhancement (KEE) techniques such as activation steering, decoding adjustments, and prompting modify inference, potentially reinforcing existing predictions or surfacing less accessible answers. Yet their evaluations centered on aggregate accuracy, which obscures diverse effects by collapsing recovered failures and disrupted successes into a single score. We introduce MAPLE, a unified framework for characterizing how KEE techniques affect parametric knowledge extraction. Its diagnostic metrics disentangle these effects at the surface, probability, and beyond-sampling levels by measuring answer recovery and disruption, gold-token rank shifts on unrecovered failures, and recovery of answers missed by fixed-budget sampling. We evaluate six KEE techniques spanning decoding, activation steering, and prompting across three model families: OLMo2, Qwen2.5, and Llama3. Our analysis reveals distinct effects and trade-offs across techniques, highlighting the value of MAPLE for understanding and comparing KEE beyond aggregate accuracy.
Oct 14, 2026

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

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.
Oct 13, 2026

Genesis Steering: Better Activation Steering Through Training Data

Activation steering controls large language models at inference time, but its effectiveness depends on the choice of reference data. We introduce genesis steering, which derives a genesis vector from contrastive examples in the model's own fine-tuning data. When these data are unavailable, we draw on the membership inference literature to identify likely training examples and construct a proxy dataset. We establish two theoretical results on optimal displacement to the decision boundary and reduced proxy-estimation bias. We evaluate genesis steering and membership-based proxy construction on Tülu-3-8B, whose fine-tuning data are publicly available, and Llama-3.1-8B-Instruct as a control. We find that safety data from Tülu-3-8B's training mixture yield stronger steering than external datasets on both models. On Tülu-3-8B, membership scores largely agree in identifying training examples, and vectors built from top-ranked examples achieve competitive performance against random subsets of the same size. In contrast, the results on Llama-3.1-8B-Instruct hint that a MIA method will only have a positive effect if the mean direction shift with MIA-filtered in-distribution data is much less severe than the direction shift with OoD data.
Oct 13, 2026

Mitigating Cross-Modal Hallucinations by Correcting Modality Misrouting in Omnimodal LLMs

Omnimodal large language models (OLLMs) are susceptible to cross-modal hallucinations under audio-visual interference. We find that, for a substantial fraction of incorrect joint predictions, the question-relevant unimodal branch can still recover the correct answer, suggesting that task-relevant evidence can remain recoverable from the relevant modality even when it is not reflected in the joint prediction. We refer to this recoverable failure pattern as modality misrouting. We propose CaRe (Conflict-aware Residual Intervention), a training-free inference-time method for identifying and correcting modality misrouting. CaRe uses a text-only modality probe to identify a candidate modality-specific counterfactual branch, and flags potential misrouting when its prediction conflicts with the original joint prediction. For such cases, CaRe computes a residual direction from the discrepancy between the counterfactual and fused representations and uses it to steer the original multimodal state toward the relevant modality evidence, while retaining both modalities in the original audio-visual inference path. The intervention strength is adapted based on the confidence of the counterfactual branch. Experiments on CMM and AVHBench across two OLLMs show that CaRe consistently outperforms the evaluated inference-time hallucination-mitigation baselines.
Oct 13, 2026

Cross-Lingual Transfer of Evil Vectors Is Directionally Asymmetric

Persona vectors are linear directions in the residual-stream activations of LLMs that encode behavioral traits and enable fine-grained control over model behavior. Among them, the evil persona vector is particularly relevant to understanding harmful behavior, jailbreak vulnerabilities, and the unlearning of harmful knowledge. Yet whether this vector reflects a genuinely cross-lingual concept, rather than an artifact of its extraction language, remains untested. We conduct a systematic evil-vector transfer study across two high-resource languages (English, Spanish) and two low-resource languages (Swahili, Telugu) on four open-source instruction-tuned LLMs, and find that transfer is directionally asymmetric. Evil vectors are harder to obtain reliably from low-resource languages, and the resulting vectors are largely ineffective at steering, even within their own language. In contrast, vectors extracted from high-resource languages transfer effectively to low-resource inputs while largely preserving general knowledge, as measured by MMLU. Geometric analysis further supports this pattern: evil vectors from the two high-resource languages are closely aligned, whereas low-resource vectors align only weakly with this shared direction. These results suggest that a shared cross-lingual evil direction exists and, once reliably extracted from a high-resource language, can be reused to steer behavior in low-resource languages without additional per-language data collection.
Oct 13, 2026

Perturbation, Not Provenance: Why Activation Steering Does Not Make a Robust Text Watermark

Linear directions in the activation space of large language models causally and universally influence model behavior. This suggests that they might serve as a basis for a robust text watermark. The idea is to steer the generation along a specific semantic direction and then recover the mark by probing the activations of the re-encoded text. We test this hypothesis by conducting a series of experiments. We vary the strength of the steering and use four different directions (formality, sentiment, complexity, and a random control) across two different model families. We also test two paraphrasing attacks: a sentence-level one and a stronger paragraph-level one. The results show that activation-steering watermarks are outperformed by a standard logit-based watermark (KGW) in terms of detection accuracy. At a matched level of fluency (within 1.15 times the original text's perplexity), KGW achieved an AUC of 0.81 against a sentence-level paraphraser, while the best steering direction only reached 0.65. A stronger paragraph-level paraphraser lowered both (0.65 for KGW and 0.59 for steering), narrowing the gap but not reversing it. On Qwen, it left the two indistinguishable but at near-chance levels for both. We further showed that the detectability provided by steering was not semantic: a random direction was just as detectable as a meaningful one. The signal was independent of which layer was probed and whether a surface feature control was used. We conclude that, in this case, semantic activation does not encode a localized, recoverable semantic mark, but rather a global, content-independent shift that can only be strongly detected when the text is degraded.
Oct 13, 2026

Beyond Bias Scores: Answerability Confounds in Activation Steering with Gradient-Constrained Direction Design (GCD)

Bias scores in language-model evaluations often conflate what a model prefers with whether it answers at all. We identify this confound in BBQ, whose ambiguous score factorizes into answerability and signed preference. Across 11 categories, 28,}824 items, and five models, contrastive steering lowers the signed ambiguous score in 11/11 categories but lowers the signed disambiguated preference score in only 8/11. Across 99 configurations, the answerability channel exceeds half of the observed score change in 17% and accounts for the entire change in 11%. In a six-category disambiguated evaluation, 20.8% of 1,500 items flip to the unknown option, where abstention is incorrect; StereoSet shows no analogous answer-rate shift. The effect reflects selection and near-orthogonality between the contrastive and preference-gradient directions. We introduce Gradient-Constrained Direction Design (GCD), an option-gradient construction with a first-order unknown-logit constraint. Under a matched Qwen-7B protocol, GCD reduces answer-rate distortion 14-fold and yields a 4-fold larger preference change, while ablations show similar gains for alternative and unconstrained option-gradient directions; nulling the unknown gradient alone is not responsible for the gain. We recommend reporting answerability alongside published scores.
Oct 12, 2026

When a Conceptor Complement Becomes Gain: Operator Geometry and Behavioral Attribution in Language Models

When a steering intervention changes a language model's behavior, which part of the intervention deserves the credit? We test this question for a target-centered, scaled conceptor complement by removing its fitted covariance filter while preserving the same target-specific center and gain. At the original aperture α=.05α=.05 and gain β=2β=2, all 21 fitted maps across three models and seven targets expand every direction at the generation intervention layers. The matrix correction is 0.093%0.093\%--7.815%7.815\% of the matched-control displacement on held-out extraction activations. In 480 paired comparisons within a 1,920-response study, retaining the matrix changes many individual texts but does not establish a consistent additional target-suppression benefit on the four-target panel. A geometric aperture sweep shows that the operating regime changes with the filter setting. Separately, an invertible soft filter preserves what an unrestricted affine readout can recover, whereas dedicated erasure methods reduce prediction by newly fitted probes. Together, the results show why a whole-intervention comparison cannot by itself attribute generated behavior to the fitted matrix: analyze the complete transformation, hold shared components fixed, and test its output consequences directly.
Oct 12, 2026

REVEAL: Reverse Engineering Agentic Latents from Frontier-Model Traces

Frontier LLM agents perform complex interactive tasks, but their inference costs motivate transferring their experience to smaller open-source agents. Distillation and latent communication typically require target-model training or access to source-model internals. We study how textual frontier-model trajectories can provide reusable local controls for a frozen agent. Inspired by protocol reverse engineering, we introduce Agentic Latents\textit{Agentic Latents}, comprising anchored states and dynamic profiles of feedback--response cycles elicited by replay in the target model. Our training-free framework, REVEAL\textbf{REVEAL} (REV\textbf{REV}erse E\textbf{E}ngineering A\textbf{A}gentic L\textbf{L}atents), uses these profiles and trajectory outcomes to match and select local events from successful and failed rollouts of the same task and source model. Their response-transition contrasts yield a target-specific steering direction, applied before each response. Within-benchmark evaluations on Terminal-Bench 2.1 and Humanity's Last Exam with tools report higher accuracy than unmodified agents across three targets. Representation analyses and behavioral case studies support reusing frontier-model experience as local activation control.
Oct 12, 2026

The Sharp Directions Are Against You: Curvature Analysis of Activation Steering

Activation steering adds a vector to a language model's internal representations at inference time. It is a lightweight alternative to fine-tuning for behavioral control. The standard construction, contrastive activation addition (CAA), is fragile, as on many inputs it shifts behavior in the wrong direction. To explain this fragility, we first study the geometry of the steering vector. We introduce steering Hessian, the matrix of second order derivative of the model's behavioral loss with respect to the steering vector. It captures the output behavior response sensitivity of the model under small perturbations of the steering vector along each direction. Its top eigenvectors are sharp directions: tiny perturbations along them can cause large, erratic, input-dependent behavioral swings. Bottom eigenvectors are flat directions: behavior is robust along them. Removing the sharp component of a steering vector approximately preserves its direction and magnitude (cosine > 0.974, norm reduction < 3% for typical settings) while reducing the vector's directional curvature by up to 12x (median 4x across 36 conditions). We evaluate on three model families (Qwen2.5-7B, Llama-3.1-8B, Mistral-7B), with four behavioral tasks, and 36 token-level conditions. Steering along sharp directions reverses the intended behavior in every condition tested, with anti-steer rates of 53-100%. Sharp directions account for only 1-5% of the CAA vector's squared norm. However, they drive a disproportionate share of the gap between the steered and unsteered next-token distributions during autoregressive generation. We term this one-line correction as sharp removal; it outperforms standard CAA in open-ended generation across all 25 conditions tested. A jailbreaking experiment further shows that sharp directions act as noise rather than signal: they disrupt coherent behavioral control regardless of steering polarity.
Oct 11, 2026

Harmonic Representations in Symbolic Music Transformers are Localized and Controllable

Transformer models trained on symbolic music have become remarkably capable. However, it is yet unclear how they represent music structure, and whether understanding their internal representations can be used to unlock methods for model control. We examine this through a case study of harmonic analysis: when a music transformer navigates a cadence, what does it internally "know" about its harmonic context, and can that knowledge be used to affect the generation? Using Aria, a LLaMA-based symbolic music transformer evaluated on Bach chorales, we find that harmonic decisions are causally localized progressively from middle layers onward. Linear probes show that mode becomes decodable in the same layer range where patching identifies growing causal effect, while key identity emerges later. We then confirm these representations are actionable for music generation control: steering via latent directions corresponding to harmonic changes reliably produces a deceptive or minor resolution, while preserving structural coherence and generation quality. This demonstrates that, at inference time, the model's internal harmonic knowledge can be directly manipulated to control the harmonic output.
Oct 11, 2026

RIVO: Lightweight Adaptive Rerouting for Deep Research Agents

Deep Research agents rely on long-horizon reasoning and iterative tool use to solve complex information-seeking tasks. However, adaptive decision-making over these trajectories remains challenging, as agents may stop prematurely when further evidence is needed, continue unnecessarily after sufficient evidence has been gathered, or fail to redirect unproductive trajectories. Existing solutions either require costly model retraining and dense turn-level supervision, or rely on heuristic control rules that are not directly optimized for downstream research utility. This motivates a lightweight, utility-aware alternative for Deep Research control. We formulate adaptive rerouting in long-horizon Deep Research as a state-conditioned trajectory correction problem. Based on this view, we introduce RIVO, a lightweight plug-in framework for adaptive trajectory correction. The agent estimates its research state from lightweight signals and determines whether trajectory redirection is needed. Once triggered, the plug-in modifies the Host token logits to steer generation toward a more promising research direction. Theoretical analysis shows that, under a value-alignment condition and bounded local curvature, a sufficiently small logit intervention can locally improve expected downstream research utility. We further construct BrowseCore, a curated dataset with a paired local corpus, for efficient and controlled Deep Research experimentation. Building on this resource, the plug-in is trained from downstream outcomes, enabling it to learn corrections that improve subsequent research and final task performance. Extensive experiments across multiple Deep Research benchmarks and control settings demonstrate the effectiveness of our method, showing that lightweight logit intervention can improve long-horizon research without retraining the underlying Host model.
Oct 11, 2026

Mind Your Own Role: Reducing Role Overstepping for Effective and Efficient Multi-Agent Collaboration via Dynamic Role Activation Steering

Multi-Agent Systems (MAS) have shown strong capability in solving complex tasks by assigning different agents specialized roles and corresponding subtasks. However, agents may overstep their assigned responsibilities and perform subtasks belonging to other roles, which can disrupt workflow, reduce system efficiency, and degrade task performance. To address this issue, we investigate role-related patterns in the model's internal activation space and uncover two key observations: (i) role instructions induce consistent directional shifts in activation space, and (ii) role-adherent and role-overstepping trajectories exhibit distinct activation patterns. Based on these observations, we propose Dynamic Role Activation Steering (DRAS), a lightweight and effective inference-time activation intervention approach to reduce role overstepping in multi-agent collaboration. Specifically, DRAS constructs a role steering vector and a lightweight role-overstepping detector for each role. During LLM generation, the detector estimates the probability of role overstepping from the current activation state, and DRAS dynamically adjusts the steering strength accordingly to guide the agent toward its assigned responsibilities. Experimental results show that DRAS effectively reduces the number of role-overstepping trajectories by up to 86.6% compared with Vanilla MAS, thereby improving task quality score by 10.0%--11.1%, achieving a 1.50×\times--1.61×\times task-completion speedup, and reducing token consumption by 26.8%--38.7%.
Oct 11, 2026

TRACE: Temporal Risk Assessment and Concept Editing for Safer Language Generation

Large language models (LLMs) can produce harmful content while generating a response. A useful safeguard should not only detect this risk during generation but also intervene while leaving safe responses largely unchanged and provide an auditable record of the internal patterns supporting its decisions. We propose TRACE (Temporal Risk Assessment and Concept Editing), a framework that tracks harmful-response risk during generation, selectively edits token activations, and leaves an auditable trace of its detection and steering decisions. TRACE learns a codebook of recurring activation patterns, with each codebook vector representing a latent concept. At inference time, it assigns each token activation to the nearest codebook vector to obtain the corresponding concept. A gated recurrent unit (GRU) processes the activations and their assigned concept features to predict completed-response harmfulness and a monotone cumulative streaming risk. During steering, activations assigned to harmful concepts are steered along directions defined by nearby benign targets. An external LLM labels the learned concepts post hoc, making detection sequences and harmful-to-benign edit records interpretable. Experiments across three generators and the WildGuard and S-Eval benchmarks show that TRACE improves streaming F1 over Kelp in five of six comparisons while remaining competitive on completed-response detection. TRACE's selective steering reduces Harmbench attack success with little change in refusal on originally safe responses. The post-hoc audit reveals recurring semantic structure in the learned concepts and distinct traces associated with harmful detection, safe generation, and activation edits.
Oct 10, 2026

FleetPilot: Fine-Grained Steering Empowers Real-World LLM Personalization

With the rapid evolution of large language models (LLMs), there is an increasing demand for effective personalization techniques that can adapt model behavior to individual user preferences. Personalized LLMs must infer a user's query-dependent preferences from limited, heterogeneous, and evolving interaction histories. Existing approaches rely primarily on in-context personalization, parametric adaptation, or personalized reward modeling, but respectively incur recurring costs in context management, model optimization and deployment, or multi-candidate generation and scoring. These limitations become particularly restrictive when user observations are sparse and the preference evidence relevant to each query varies across contexts and over time. To address these challenges, we propose \modelname, a fine-grained and instance-tailored steering framework that dynamically generates sample-level interference vectors from user data and injects them into the model's forward pass for personalized adaptation. Our approach introduces: match contrast that produces user-relevant signals, fine-grained steering component that captures nuanced signals by hooking activations from attention and MLP layers, and query-conditioned aggregation module that synthesizes these signals into contextually relevant enhancements. Extensive experiments across diverse scenarios, including short-to-long text generation, and web function calling, validate the effectiveness and compatibility of \modelname. Results show that our method consistently brings significant performance gains, and show extreme superiority in fast-shifting and data-sparse environments. Additionally, \modelname rids us of training cycles and incurs light overhead of storage or inference latency. These make \modelname a strong and real-world deployable method. Implementation is available \footnote{https://anonymous.4open.science/r/FleetPilot}.
Oct 10, 2026

What Survives When Activation Steering Is Used as Memory

Activation steering has recently been used as a memory for skills: past experiences of an agent are stored as steering vectors and retrieved later. Facts and preferences about the user are assumed to need text instead. We test that assumption with a training-free activation memory beside a frozen LLM. The change an experience causes in late-layer states is stored as a shift, filed under a whitened key for the conversation's gist, and added back in a later session only when the gist matches. We argue that such a shift is a fixed average of an effect that attention, in context, recomputes for every question, so only question-independent content can survive. In a pre-registered study on Qwen3.5-2B and 9B, with judges from four model families, the memory is selective (about 98% of unrelated replies are unchanged), but facts are stored as the word to say, in the last layers only, and are almost never used (at most 15% clean use, against 96-98% in context). Leanings transfer only in a narrow strength window. Most strikingly, the reference subtracted when writing a preference decides whether its concept or its direction survives: unless the opposite is subtracted, a stored dislike is recalled as a like.
Oct 7, 2026

Steerspeech: Activation Steering For Emotion Control In Generated Speech

Pretrained text-to-speech (TTS) models can generate expressive speech, but reliable inference-time emotion control remains challenging: prompts and reference audio offer coarse, inconsistent control, whereas specialized conditioning and model adaptation require costly training. We present SteerSpeech, a lightweight activation-steering framework that controls emotion by injecting steering vectors into hidden activations. For each target emotion we train a lightweight low-rank transform, using a multi-expert objective that encourages monotonic emotion control while preserving speaker identity and linguistic content, constraining steering drift, and keeping the TTS backbone frozen. To optimize through discrete speech tokens, we introduce a two-pass generation-and-replay pipeline using a straight-through estimator to backpropagate expert supervision through sampled tokens. At inference, a target-emotion steering direction is optimized with its respective transform and injected into the base TTS model. Objective and subjective evaluations with Qwen3-TTS across seen, unseen, and accented speakers show stronger continuous emotion control with limited speaker and content degradation. SteerSpeech achieves 1.08x-7.12x baseline target-emotion scores and for a representative emotion subjectively, it receives 78.1%-96.8% intensity preference and 1.43x-1.46x speaker-identity preservation at high steering strengths.
Oct 7, 2026

Nullify: Null-Space Activation Steering for Training-Free LLM Unlearning

Large Language Models (LLMs) inevitably internalize substantial amounts of sensitive or private information during pre-training, while LLM unlearning aims to selectively erase specific knowledge to prevent privacy leakage with minimal loss of model utility. However, existing methods struggle to balance forget quality with utility, and typically incur substantial computational costs due to parameter fine-tuning. To address this, we propose Nullify, a training-free, non-destructive activation steering method for LLM unlearning. Nullify employs steering vectors during inference to redirect privacy-related activations away from their memorized answers, while satisfying a null-space constraint that leaves retained-query activations essentially unaffected to maintain utility. Evaluations on TOFU and MUSE show that Nullify matches or surpasses established baselines in forget quality while achieving near-lossless preservation of model utility. By avoiding weight updates entirely, Nullify serves as an efficient, plug-and-play inference-time intervention framework.
Oct 7, 2026

PatchBench: Measuring Collateral Damage in Activation Patching

An LLM safety patch can pass a benchmark while still being a poor repair. This risk is especially acute for jailbreak repairs, where the goal is to correct a specific unsafe behaviour without changing unrelated behaviours. A patch may block exact evaluation prompts yet fail on close harmful variants, or suppress harmful behaviour by over-refusing benign prompts that share its wording or structure. Existing protocols primarily test whether models can be broken, while aggregate metrics (attack success, refusal rates, global capability) cannot distinguish selective repairs from broader local suppression. To address this gap, we introduce PatchBench, a benchmark of empirically observed model-specific jailbreak failures inducing actionable harmful answers. Starting from 27,870 prompts from 37 public datasets, we curate 15,314 English prompts and query 8 open-source instruction-tuned models. Combining WildGuard filtering, pairwise Elo ranking, and manual verification, we retain a curated bank of 400 high-confidence jailbreak failures. We further introduce PatchBench-Local, an evaluation protocol testing whether a patch is behaviourally precise. For each harmful source prompt, PatchBench-Local generates three families of local neighbours: harmful variants preserving malicious intent, benign prompts with matched structure, and benign prompts reusing key harmful terms. It evaluates harmful-neighbour correction and benign-neighbour preservation, distinguishing selective repair from broader local suppression. Evaluating four activation steering methods with PatchBench-Local and MMLU shows that global capability can remain nearly unchanged while local benign regressions are severe, confirming aggregate metrics miss important collateral damage. PatchBench-Local provides a more precise basis for developing and comparing jailbreak repair methods.
Oct 7, 2026

The Attribution Blind Spot: Layerwise Trajectory Diagnostics for Source Reliance in Retrieval-Augmented Language Models

A retrieval-augmented model can match a document without relying on it. Controlled knowledge conflicts make source choice observable and let us ask a second question that prediction alone cannot answer: which internal-state properties define useful intervention directions? We study paired hidden-state changes with Latent Trajectory Shift (LTS), a signed projection onto a training-fitted first principal component (PC1), and keep verified training exposure separate from behavioral source choice. Across the evaluated conflicts, state-change magnitude is often the stronger predictor, whereas signed PC1 is the stronger selective controller: equal-norm interventions change source preference while better preserving non-target behavior, and the frozen direction transfers across the tested datasets and aligned model pairs. A same-system OLMo study further combines positive choice and control results with inconclusive exposure detection at the achieved power. The central result is a separation: representations that diagnose what a model will choose need not be the representations that best control that choice.
Oct 6, 2026

Steering Follows Geometry, Not Labels: Emotion Directions in a Full-Duplex Speech Model

Full-duplex voice agents need to modulate emotion and delivery during real-time conversations, when de-escalating a complaint, carrying urgency in dispatch, softening a clinical result. Emotion and delivery control is well studied for TTS and turn based models through prompt-conditioned synthesis, reference-conditioned synthesis and activation steering; PersonaPlex controls identity in a duplex model but not affect. We study emotion steering in Moshi, a fully open sourced full-duplex speech language model, across four emotions, using mean-difference activation steering, which costs only a few vector additions per frame and no retraining. We show that emotion is linearly decodable from Moshi's residual stream, but activation steering is only partially achievable, and unevenly so; as happy, angry and surprise steer towards a shared direction while sad is distinctly steerable. We also show that the shared component across the three emotions cannot simply be projected away from all the emotions equally.
Oct 6, 2026

Does Steering Break Your Model? A Multi-Dimensional Evaluation Suite for LLM Steering Methods

Activation steering provides a lightweight and flexible way to control large language model (LLM) behavior. However, effective steering requires more than inducing the intended behavior: it should also limit unintended changes and remain robust across inputs and training data. Existing evaluations cover these dimensions only in fragments. As a result, the trade-offs between efficacy and side effects have not been systematically characterized. We introduce SteerScope, a two-axis, multi-dimensional evaluation suite that jointly characterizes steering outcomes and method properties through 15 metrics. We score target efficacy and side effects on language quality, task capabilities, and safety and reliability, and further assess generalization and data dependence through steering-specific metrics for sample efficiency and sample sensitivity. Rather than comparing methods at a single operating point, we characterize the trade-offs between efficacy and side effects. Under matched models, tasks, and evaluation protocols, we benchmark 23 methods spanning 4 families, including prompting, LoRA, and SFT as baseline methods, and release the suite as an extensible codebase. We find that current activation steering methods do not yet surpass the Prompt Steering baseline in their overall balance between steering efficacy and side effects: across both model scales, no evaluated activation steering method achieves higher efficacy without incurring greater composite side effects. We further uncover a consistent coupling between steering efficacy and side effects. Under OOD prompts, target efficacy is often preserved, whereas side effects tend to become more pronounced, particularly through declines in instruction relevance and fluency. Methods also exhibit sharply different sample-efficiency profiles.
Oct 6, 2026

Loud and Clear: Dynamic Activation Steering for Improving Speech Intelligibility in Noisy Environments

Speech becomes less intelligible in noisy environments, and humans naturally adapt their voice to compensate. Inspired by this behavior, we investigate whether a text-to-speech (TTS) model can be guided to produce more intelligible speech using activation steering, without retraining. We focus on two characteristics of the Lombard effect: increased vocal effort and hyper-articulation. We introduce a prompt-relative steering mechanism that prevents steering effects from accumulating during generation while allowing their strength to be adjusted dynamically. Across seen and unseen speakers and multiple languages, our method produces systematic changes in Lombard-related acoustic features, preserves speaker similarity (89-95%), and reduces WER under background noise by 7-22% at 1 dB SNR. These results show that pretrained TTS models can be dynamically controlled to generate more intelligible speech without retraining.
Oct 5, 2026

COMPASS: Finding Where Reasoning Lives in Language Models

Explicitly eliciting reasoning substantially improves LLM performance. Existing approaches require a predefined characterization of reasoning, whether through CoT prompt design, contrastive CoT directions, or via SAE derived reasoning features. For mathematical reasoning with verifiable answers, we show that a much simpler signal suffices, which is the correctness of the model's own direct answer attempts. This signal yields a latent direction that elicits reasoning. This direction is decodable within the activations of most attention heads, but only a small subset of them can be effectively intervened. We introduce COMPASS, an inference-time steering method that identifies these heads using a logit-space attribution score and steers their activations along the correctness direction, requiring only per-head activation statistics. Across three model families and multiple math benchmarks, COMPASS outperforms the activation-steering baselines we compare against, improves GSM8K accuracy by 16 percentage points on average, and approaches CoT accuracy with 20-70% fewer generated tokens. Interventions transfer without re-fitting to unseen benchmarks, and ablations show that both the correctness direction and the small set of heads carrying it are necessary, with the effect concentrated in remarkably few heads.
Oct 5, 2026

Defense-in-Depth for LLMs: Evaluating Memory Gates Against Activation-Induced and Memory-Induced Sycophancy

Long-term memory allows Large Language Models (LLMs) to maintain personalized context across interactions, but retrieved user history can induce memory-induced sycophancy, causing models to favor stored user beliefs over objective evidence. Existing defenses primarily operate on retrieved context and are rarely evaluated jointly with internal behavioral bias. We introduce a 2×22 \times 2 defense-in-depth framework separating internal activation steering from external memory handling. We extract sycophancy steering directions from 100 paired prompts and evaluate four open-weight models across 10 steering coefficients and five memory-defense configurations on MemSyco-Bench (answers for all 1,550 items; defense conditions judged on a fixed 250-item subsample), with three LLM judges. Three of the five configurations are new (rewriting every memory, a Router Gate that keeps, rewrites, or drops each memory, and dropping all memory); the other two are MemSyco's baselines. Selective Router Gate filtering preserves substantially more of MemSyco's average accuracy than complete memory removal, and this separation persists when the models are steered toward sycophancy. On Llama 3.1 8B with Router Gate, mild inverse steering (α=−1.5α= -1.5) lowers judge-averaged sycophancy from 35.80% to 31.32% while average accuracy moves from 43.99% to 43.31%; this reduction has the same direction under all three judges but is not statistically significant (paired p=0.08p = 0.08 to 0.630.63 on 149 items). External memory filtering is the part of the design that holds up; our data do not show that inverse steering adds to it.
Oct 5, 2026

Steering by Influence: Curvature Aware Data Weighting for Activation Steering

Inference-time steering offers cheap, fine-grained control over a language model's outputs by estimating a concept's representation in activation space and shifting activations towards it. Existing methods build these representations from activation averages over contrastive datasets. These averages incorporate unrelated concepts and noise, and are dominated by a few tokens, meaning the activation transport encodes token-level rather than thematic concepts. In this work, we steer towards examples that most express a concept thematically, rather than towards an expectation over all. We identify these examples using influence functions, which estimate how much each data point contributes to a model's representation of a concept. Unlike simple model activation similarity, they incorporate the curvature of the model's loss landscape, allowing them to capture concept-relevant relationships beyond superficial token-level similarity. We then propose influence-weighted activation transport, which uses optimal transport to steer activations of non-concept text towards those of concept text, weighting concept examples by their influence scores. We evaluate on toxicity suppression (Jigsaw), object-based concept induction (OneSec) and truthfulness induction (TruthfulQA), outperforming existing activation-transport baselines. We track capability after steering using perplexity and MMLU accuracy, finding that our method improves steering while largely preserving model quality. We further show that influence functions capture concept-relevant information that activation-based methods miss with the two approaches ranking data points significantly differently. Together, these results demonstrate the value of curvature-aware influence information for activation steering.
Oct 5, 2026

TrustMI: Causally controlling how assistants trust their users

Large Language Model (LLM) assistants routinely decide whether they can trust users and third parties whose competence, intentions, and integrity they cannot verify. This uncertainty matters for safety, as trusting the wrong party can lead an agent to comply with harmful requests or act on malicious instructions encountered during tool use. To study this problem, we define trust as an assistant's willingness to accept vulnerability to the actions of another party and ask whether such behavior can be causally controlled through model activations. We build 2,000 contrastive conversations spanning ability, benevolence, and integrity, where paired responses complete the same request but differ in whether the assistant trusts the user. From these pairs, we learn steering matrices while keeping the model parameters frozen and test them across six instruction-tuned models from three families, finding that steering changes trust decisions monotonically in both directions. We then ask whether this effect extends to several safety-related agent settings involving harmful requests, prompt injections, and insider threats, while using benign-task and reasoning as controls. Our findings provide evidence that trust in the user can be causally controlled along linear directions in model activations and provide a way to study how trust shapes safety-relevant behavior in language models.
Oct 5, 2026

Noise Out, Bias In: Targeted Bias Injection in Diffusion Language Models via Closed-Loop Activation Steering

Masked diffusion language models (dLLMs) generate text by iteratively denoising masked positions, re-predicting each token multiple times before it is committed. An autoregressive decoder exposes an answer's distribution once, at the step that commits it; a dLLM exposes it at every denoising step before commitment, and we show that an adversary can exploit this. Since an answer remains open to revision over many denoising steps, an adversary with access to internal activations can watch how likely the model is to produce a chosen answer and adjust the intervention accordingly. Building on this observation, we study targeted bias injection, an attack that steers a frozen dLLM toward a demographic answer selected by the adversary. The attack uses a simple proportional-integral (PI) controller that tracks the target-answer probability during denoising and adapts the strength of a steering vector on the fly. On ambiguous BBQ questions where the correct answer is abstention, our attack raises LLaDA-8B-Instruct's preference for the targeted group from 1.8 to 16.7 percentage points, more than three times the strongest fixed-strength steering baseline, and on SocialStigmaQA it raises the selection of stigmatizing answers from 17.6% to 58.1%. Fitted to other demographic targets, the same attack shifts answers by up to 37 percentage points, and each attack takes about 40 minutes on one GPU. On the primary target, feedback is what makes the attack work: constant steering at the same average strength over the token-committing steps produces a far smaller shift while corrupting nearly three times as many outputs, and a constant strength set separately for each example still falls well short. Our findings identify the denoising trajectory as a new control channel in dLLMs and call for bias audits that examine the serving stack rather than the frozen model alone.
Oct 4, 2026

Usage-Modulated Sentiment Representations in Large Language Models

Prior work suggests that sentiment can often be captured by approximately linear directions in LLM activation spaces, but a single direction may not fully capture sentiment representations. In natural communication, sentiment is shaped not only by polarity but also by usage factors, such as tone and audience adaptation. We test whether these factors systematically modulate sentiment representations beyond a shared sentiment direction. We construct a controlled paired dataset that holds event content fixed while varying sentiment polarity and usage factors, and analyze Llama, Mistral, and Gemma. We identify a shared sentiment direction, remove it, and test the residual structure through erasure and generation-time tone steering. Across models, the shared direction is robust (median cosine 0.953-0.975), yet removing it leaves 0.833-0.909 of the original positive-negative representation-difference norm. The residuals contain compact, reproducible usage-conditioned structure. Targeted erasure weakens held-out usage metrics more than random and label-shuffled controls. On Llama, outputs steered along residualized tone components are preferred in 92.8% of blind target-tone comparisons while preserving the requested sentiment polarity in 98.7% of evaluated outputs.
Oct 4, 2026

EmoRSS: Mitigating Emotion-Induced Over-Refusal in Large Language Models

Emotional expression can influence the safety decisions of large language models (LLMs), offering a potential avenue for improving safety alignment. Existing studies have mainly focused on how emotional expressions facilitate attacks under harmful requests, while overlooking their effects on benign requests. We find that emotional expression can also systematically increase refusal tendencies on benign requests, leading to unnecessary over-refusal. Based on this observation, we propose emotion-guided refusal subspace steering (EmoRSS), an activation-steering method that mitigates emotion-induced over-refusal while preserving refusal behaviour on harmful requests. Specifically, we first identify a refusal-sensitive layer using layer-wise linear probes and construct a refusal subspace from sparse autoencoder (SAE) features aligned with the probe direction. Next, we use paired regular and emotional requests with the same queries to estimate the mean activation shift in the features defining the refusal subspace. Finally, we decode this shift into an activation intervention vector and apply it in the reverse refusal direction during inference, without updating the backbone parameters. Experiments on two LLMs show that, when requests contain emotional expressions, our method achieves a more favourable trade-off between refusing harmful requests and answering benign ones than prior over-refusal mitigation baselines, while better preserving general task performance.