cs.LGNov 18, 2026

A Rank-One Bypass of Fixed Linear Erasure

Authors: Jai Kumar Sharma, Vanshaj Khattar

Organizations: Virginia Polytechnic Institute and State University

Abstract

Linear concept erasure can make a target unreadable to a specified family of linear classifiers applied to a model's current internal features. In an open-weight release, however, an attacker can modify the layers that produce those features. We study a fixed affine erasure layer whose upstream model is allowed to change. If the layer transmits even one direction visible to an allowed downstream classifier, the attacker can write a target score into that direction and carry it through the layer. We show that this condition is also necessary, and that for a nonconstant linear target score the required change is a rank-one affine update. Across six CIFAR-10 class-pair tasks, the route restores nearly all pre-erasure target performance and can leave a deployed benign linear head's scores unchanged to numerical tolerance. We then price the attack: with eight target labels and 2,000 unlabelled calibration examples, a 10% mean relative feature edit yields 79.8% test accuracy, which is 91.2% of the fitted score's gain over chance; tighter edit limits sharply reduce success. Linear guardedness therefore remains a valid guarantee about the current representation, but it is not by itself a tamper-resistance guarantee for an editable open-weight model.

Figures & tables

Appendix figures & tables6 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Mar 8, 2026

Obliviator Reveals the Cost of Nonlinear Guardedness in Concept Erasure

Concept erasure aims to remove unwanted attributes, such as social or demographic factors, from learned representations, while preserving their task-relevant utility. While the goal of concept erasure is protection against all adversaries, existing methods remain vulnerable to nonlinear ones. This vulnerability arises from their failure to fully capture the complex, nonlinear statistical dependencies between learned representations and unwanted attributes. Moreover, although the existence of a trade-off between utility and erasure is expected, its progression during the erasure process, i.e., the cost of erasure, remains unstudied. In this work, we introduce Obliviator, a post-hoc erasure method designed to fully capture nonlinear statistical dependencies. We formulate erasure from a functional perspective, leading to an optimization problem involving a composition of kernels that lacks a closed-form solution. Instead of solving this problem in a single shot, we adopt an iterative approach that gradually morphs the feature space to achieve a more utility-preserving erasure. Unlike prior methods, Obliviator guards unwanted attribute against nonlinear adversaries. Our gradual approach quantifies the cost of nonlinear guardedness and reveals the dynamics between attribute protection and utility-preservation over the course of erasure. The utility-erasure trade-off curves obtained by Obliviator outperform the baselines and demonstrate its strong generalizability: its erasure becomes more utility-preserving when applied to the better-disentangled representations learned by more capable models.
Oct 23, 2026

Principled Nonlinear Concept Erasure and Steering

Learned representations often encode sensitive attributes, such as gender or race, that may be undesirable or unsafe to retain. Concept erasure aims to remove this information while preserving task-relevant content. Existing methods remain incomplete: linear methods provide guarantees only against linear probes, while nonlinear methods achieve stronger empirical erasure but lack formal guarantees and often sacrifice utility. In this work, we introduce a principled conditional autoencoder framework for nonlinear concept erasure. We follow the principle that erasure should remove only the sensitive attribute and nothing else, and we show that, under the stated assumptions, perfect reconstruction together with population-level independence is sufficient for perfect nonlinear guardedness, with preservation guarantees in the imperfect-reconstruction regime. Empirically, we attain a superior utility-erasure trade-off across multiple datasets, sensitive attributes, and language models, surpassing even the supervised variants of competing methods despite using no target-task labels, while being simpler to optimize and converging substantially faster. Beyond erasure, our framework yields concept conversion (representation steering) as a by-product: by changing only the target concept at inference time, it produces representations corresponding to different concept values - at no additional training cost - a capability no prior erasure method provides, and on which we likewise demonstrate strong empirical performance.
Sep 5, 2026

Bait-and-Recover: Poisoning Internal Refusal Signals to Defend LLMs against White-Box Editing Jailbreaks

Open-weight large language models face a low-cost white-box threat from representation engineering attacks. Attackers can estimate refusal directions and search for projection-matrix edits that suppress safety alignment while preserving general capabilities, within minutes on a single GPU and without gradient-based training. We propose Bait-and-Recover, a weight-level defense that places a bait adapter where attackers read activations and a paired recovery adapter at the subsequent layer. Trained via gradient routing, this decouples the observation path from the behavior path. By actively poisoning the residual signal used for measurement, Bait-and-Recover disrupts the attacker's edit search, while the recovery layer restores clean downstream computation. Across four open-weight models, our defense raises the minimum refusal rate against white-box edit searches from 16.25% to 71.75% under a strict behavior-preservation budget (KL <= 0.10), with negligible impact on general benchmarks. By invalidating the core measurement assumption of these attacks, observation-path poisoning offers a practical complement to behavior-level safety training.