[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82937-en":3,"doc-seo-82937-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},82937,1099514068035,"Ezra","https://ap-avatar.wpscdn.com/davatar_276721f389ce27ea32af1340a28f341c",8,"Research & Report","Faithfulness to Refusal: A Causal Audit of Neuron Selectors","Attribution scores increasingly guide neuron-row choices in LLM workflows for pruning, interpretability, and safety editing, yet causal validity is seldom directly tested. The work presents paired causal audits using one-shot neuron-row zeroing. At the language-modeling level, attribution-based selectors identify dispensable rows far better than activation- and magnitude baselines across multiple LLMs. At the behavior level, contrastive harmful-versus-benign signals show that attributed rows install refusal on hate and crime while keeping benign over-refusal low. Rank stability can still fail; refusal is supported by a redundant, subspace-level mechanism.","arXiv :2607 .05355v 1 [ cs .CL] 6 Jul 2026  \nAnanth Eswar∗ †, Pratinav Seth∗,  \nUtsav Avaiya, Vinay Kumar Sankarapu  \nLexsi Labs  \n[pratinav.seth@lexsi.ai](pratinav.seth@lexsi.ai)  \nAbstract  \nAttribution scores increasingly identify which neuron rows of a language model matter for applications such as pruning, interpretability, and editing for safety, yet whether they identify causally important rows is rarely tested directly. We address this with two paired audits built on one-shot neuron-row zeroing. We first auditselectors at the language-modeling level: attribution methods substantially outperform activation-and magnitude-based baselines at identifying dispensable rows across five LLMs. We then adapt the same intervention into a behavior test by driving it with a contrastive harmful-versus-benign signal; the attributed rows are sufficient to install refusal on hate and crime while keeping benign over-refusal low and preserving language model fluency, and specific in that layer-matched random controls at the same depths fail. Highly rank-stable selectors can be among the least causally valid. Refusal moreover lives in a redundant subspace, where different attribution methods install it through largely disjoint row sets, so the recovered edit is one realization of a sufficient set rather than a unique mechanism. Together, these findings show that rank-stability proxies miss the kinds of selector failures a direct causal audit can surface.  \nKeywords: neuron attribution, structured pruning, mechanistic interpretability, refusal editing, selector faithfulness, LRP, integrated gradients, LLM safety  \n1 Introduction  \nAttribution scores now drive three separate workflows: pruning methods remove parameters by importance score [1, 2, 3], mechanistic-interpretability papers identify behavioral circuits via attribution [4] and safety-editing methods suppressor install behaviors by ablating attributed components [5, 6] . All three rest on the same selector-faithfulness assumption: that the scoring function (LRP, IG, Wanda, magnitude, or any other) faithfully separates important from unimportant components. The assumption fails silently. A pruning recipe wired to an unfaithful selector can still hit a perplexity (PPL) target after recovery, and a refusal edit can saturate calibration-set refusal yet generalize unsafely on held-out red-team prompts, because the standard audit artifacts (PPL, rank correlation, calibration accuracy) are blind to theselector’s failure mode.  \nA direct causal test is missing. It would intervene on the selector’s output: zero the rows it ranks dispensable, measure the damage, and do this under conditions where behavioral change can only come from the selector, not from surrounding pipeline machinery. The test must also run at two levels of difficulty, because passing the language-modeling level (does the selector pick generally important neurons?) does not entail passing the behavior-specific level (does it pick the neurons that carry a target behavior?) .  \n∗Equal contribution; co-first authors.†Work done while at Lexsi Labs.  \nFaithfulness to Refusal: A Causal Audit of Neuron Selectors  \nApproach. We use neuron-row zeroing: a one-shot, retraining-free ablation of individual output rows of a transformer’s Linear projections (attention Q/K/V/O or MLP gate/up/down) . Unlike DPO, activation steering, or representationengineering edits, it adds no fine-tuning step and no inference-time hooks, so any behavioral change traces to the rows the selector ranks dispensable rather than to surrounding pipeline machinery.  \nAt the LM level we run Least-Relevant-First/Most-Relevant-First (LeRF/MoRF) sweeps for seven selectors (Random, Magnitude, Wanda, MeanActivation, LRP, IG, and their Borda consensus CONSENSUS-2) across five base models (LLaMA-3 .2-1B/3B, LLaMA-3 . 1-8B, Qwen3-8B, Gemma-3-12B) . At the behavior level we drive attribution with a contrastive harmful-versus-benign signal on matched CAST [5] pairs: single","cbCaie6MZVQ5uYgd","https://ap.wps.com/l/cbCaie6MZVQ5uYgd","pdf",1175965,1,41,"English","en",105,"# Introduction\n## Approach\n## Contributions\n# Related Work\n## Attribution in LLMs: Methods and Faithfu","[{\"question\":\"Why is selector faithfulness an issue for current attribution-based workflows?\",\"answer\":\"The scoring functions may separate important from unimportant components incorrectly, and standard evaluation artifacts can miss the specific failure mode. This can lead pruning or refusal edits to look acceptable while generalizing unsafely or missing causal targets.\"},{\"question\":\"How does the paper test causal faithfulness of neuron-row selectors?\",\"answer\":\"It uses one-shot neuron-row zeroing to ablate the transformer output rows ranked as dispensable by a selector. By avoiding fine-tuning and inference-time hooks, behavioral changes can be attributed to the intervened rows.\"},{\"question\":\"What happens when selectors are judged only by rank stability?\",\"answer\":\"The most rank-stable selectors are reported to be the least causally faithful. At the behavior level, rank-stable selectors can even decrease refusal, showing that rank-stability proxies miss failures revealed by causal audits.\"}]",1784184141,103,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"faithfulness-to-refusal-a-causal-audit-of-neuron-selectors","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/faithfulness-to-refusal-a-causal-audit-of-neuron-selectors/82937/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"Why is selector faithfulness an issue for current attribution-based workflows?","Question",{"text":75,"@type":76},"The scoring functions may separate important from unimportant components incorrectly, and standard evaluation artifacts can miss the specific failure mode. This can lead pruning or refusal edits to look acceptable while generalizing unsafely or missing causal targets.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does the paper test causal faithfulness of neuron-row selectors?",{"text":80,"@type":76},"It uses one-shot neuron-row zeroing to ablate the transformer output rows ranked as dispensable by a selector. By avoiding fine-tuning and inference-time hooks, behavioral changes can be attributed to the intervened rows.",{"name":82,"@type":73,"acceptedAnswer":83},"What happens when selectors are judged only by rank stability?",{"text":84,"@type":76},"The most rank-stable selectors are reported to be the least causally faithful. At the behavior level, rank-stable selectors can even decrease refusal, showing that rank-stability proxies miss failures revealed by causal audits.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":45,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":45,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]