[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82224-en":3,"doc-seo-82224-105":30,"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":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},82224,1374391974468,"Eden","https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0",8,"Research & Report","Present but Rescaled Chat-to-Agent Transfer of Additive Activation Steering","Additive activation steering calibrates a scaled residual-stream direction during generation, traditionally in single-turn chat, yet targeted models increasingly run as tool-using ReAct agents. The study provides the first systematic transfer analysis from chat to agents using matched-information designs with representation readouts, transcript re-encoding to avoid KV-cache contamination, and direction-specific controls. Results show transfer occurs but is rescaled: injected directions reach late layers near full strength, while behavioral coupling varies by model and context, yielding unpredictable safety implications for agentic refusal-bypass amplification.","Present but Rescaled: Chat-to-Agent Transfer of Additive Activation Steering  \nLucas Pinto  \nIndependent Researcher  \n[lcsdgpinto@gmail.com](lcsdgpinto@gmail.com)  \narXiv :2607 .09 156v 1 [ cs .LG] 10 Jul 2026  \nAbstract  \nAdditive activation steering (injecting a scaled residual-stream direction during generation) is calibrated almost entirely in single-turn chat, yet the models it targets are increasingly deployed as tool-using ReAct agents. We present the first systematic chat-to-agent transfer study of additive steering, coupling behavioral measurement with a representation read-out in a matched-information design: the same items rendered as plain chat or as a ReAct tool-use episode, with matched-norm random-direction controls and the transcript re-encoded every turn to exclude KV-cache contamination. Transfer is real but rescaled, and the right description is a dissociation: the injected direction reaches the late layers at near-full strength in every setting and model tested (install-site agent-over-chat ratios 0.83–1.16 across three families), while the behavioral coupling is reset per model and context. On Qwen2.5-7B a refusal bypass vector amplifies in the agent (T = 1 .45, CI [1.20, 1.78], N = 300); across a powered uniform-protocol distribution the coupling spans amplification (Gemma-2-9BT = 2 .00) to attenuation (Yi-1 .5-9B T = 0 .43, CI [0 .29, 0.60]), with no universal constant and a single clean attenuator against a universal sign. Directional ablation of the same axis does not amplify (T = 0 .93, CI including 1) while additive injection amplifies (T = 1 .50), a 20.1-point gain difference (CI [13.4, 26.8]) that identifies an additive-specific mechanism. Two pre-registered instruments converge to localize therescaling to the ReAct format scaffold, before any tool observation, rather than to the observation boundary where a dilution account would predict it. The safety implication is immediate and unpredictable: agentic deployment amplifies steering-based refusal bypass by up to 2.00× on some models while others attenuate, so a deployment cannot assume a given model is safe under additive steering.  \nIntroduction  \nAdditive activation steering extracts a direction in a model’s residual stream by contrasting activations on trait-exhibiting and trait-suppressing completions, then injects a scaled copy of that direction during generation to shift behavior in a graded, interpretable way (Zou et al. 2023; Turner et al. 2023) . The technique needs no weight updates, is cheap to apply, and has been demonstrated across refusal, sycophancy, honesty, and a wide range of persona traits (Arditi et al.  \nCopyright © 2027, Association for the Advancement of Artificial Intelligence ([www.aaai.org](www.aaai.org)). All rights reserved.  \n2024; Panickssery et al. 2024) . It is increasingly proposed as a deployment-time control and monitoring primitive for safety applications.  \nAlmost all of this evidence comes from single-turn chat. The models being steered are increasingly deployed as toolusing ReAct agents: they plan over multiple turns, call external tools, read returned observations, and commit to actions across an extended context. An agent episode differs from a chat reply in ways that are not obviously neutral for an additive control method: the context is longer and its residualstream norm grows each turn, the model is rendered into a structured ReAct format rather than a free-form reply, real tool observations inject ground-truth content that can contradict a steered disposition, and behavioral commitments are often made in a planning step rather than the final naturallanguage token.  \nWhether a chat-calibrated additive control keeps its behavioral grip once the same model runs as a ReAct agent is open and practically important.  \nWhy three nearby results do not answer it. First, refusaldirection ablation (a destructive weight edit) is known to transfer to agents (Lermen, Dziemian, and Pimpale 2024): a refusal-ablated mod","cbCaipL8XS6MJhdl","https://ap.wps.com/l/cbCaipL8XS6MJhdl","pdf",390317,3,1,12,"English","en",105,"# Abstract\n# Introduction\n## Motivation and gap\n## Why related results do not answer the question\n# Approach\n## Matched-information rung ladder\n## Representation and behavioral instrumentation\n## Direction-specificity controls\n# Headline finding: dissociation","[{\"question\":\"What question does the study address about additive activation steering?\",\"answer\":\"It tests whether steering calibrated in single-turn chat maintains its behavioral effect when the same model is deployed as a multi-turn, tool-using ReAct agent.\"},{\"question\":\"How does the experiment ensure the representation measurement is not confounded by KV-cache contamination?\",\"answer\":\"The transcript is re-encoded from scratch every turn by construction, preventing KV-cache effects from altering the steering-related mechanism across agent steps.\"},{\"question\":\"What are the main findings about transfer from chat to agents?\",\"answer\":\"The injected direction survives the transfer at near-full strength, but the behavioral coupling becomes rescaled and model-dependent, with examples of amplification for some models and attenuation for others.\"}]",1784178959,30,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"present-but-rescaled-chat-to-agent-transfer-of-additive-activation-steering","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":20},"https://docshare.wps.com/document/research-report/",{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/present-but-rescaled-chat-to-agent-transfer-of-additive-activation-steering/82224/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-21","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},"What question does the study address about additive activation steering?","Question",{"text":75,"@type":76},"It tests whether steering calibrated in single-turn chat maintains its behavioral effect when the same model is deployed as a multi-turn, tool-using ReAct agent.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does the experiment ensure the representation measurement is not confounded by KV-cache contamination?",{"text":80,"@type":76},"The transcript is re-encoded from scratch every turn by construction, preventing KV-cache effects from altering the steering-related mechanism across agent steps.",{"name":82,"@type":73,"acceptedAnswer":83},"What are the main findings about transfer from chat to agents?",{"text":84,"@type":76},"The injected direction survives the transfer at near-full strength, but the behavioral coupling becomes rescaled and model-dependent, with examples of amplification for some models and attenuation for others.","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":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,122,127,130,134],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":29,"slug":121},"research-report",{"id":123,"doc_module":4,"doc_module_name":46,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":46,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":46,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":46,"category_name":136,"show_sort_weight":106,"slug":137},19,"General","general"]