Model faithfulness is closely tied to environmental factors, and faithfulness/mechinter/safety research should account for that!</p>\n","updatedAt":"2026-09-01T10:22:26.274Z","author":{"_id":"61001311e043e15c13412d30","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/61001311e043e15c13412d30/6yAbTweYR16XtxMBEyOWl.png","fullname":"Pasquale Minervini","name":"pminervini","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":63,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9303843379020691},"editors":["pminervini"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/61001311e043e15c13412d30/6yAbTweYR16XtxMBEyOWl.png"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.29464","authors":[{"_id":"6a96a6a243140794820a76d4","name":"Aryo Pradipta Gema","hidden":false},{"_id":"6a96a6a243140794820a76d5","name":"Neel Rajani","hidden":false},{"_id":"6a96a6a243140794820a76d6","name":"Rohit Saxena","hidden":false},{"_id":"6a96a6a243140794820a76d7","name":"Wai-Chung Kwan","hidden":false},{"_id":"6a96a6a243140794820a76d8","name":"Pasquale Minervini","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/61001311e043e15c13412d30/cnkZepaXk2uU1R-B39hjx.png"],"publishedAt":"2026-08-29T00:00:00.000Z","submittedOnDailyAt":"2026-09-01T00:00:00.000Z","title":"Chain-of-Thought Faithfulness of Reasoning Models Varies with Where and How Preference Cues Are Delivered","submittedOnDailyBy":{"_id":"61001311e043e15c13412d30","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/61001311e043e15c13412d30/6yAbTweYR16XtxMBEyOWl.png","isPro":false,"fullname":"Pasquale Minervini","user":"pminervini","type":"user","name":"pminervini"},"summary":"Chain-of-thought (CoT) monitoring assumes that reasoning traces faithfully record the information that shapes a model's answer. Existing faithfulness tests often place explicit bias cues in the user message, while agents may encounter preferences through tool returns or raw artifacts. We introduce FACE-Eval (Faithful Attribution of Cue Effects Evaluation), a 5,100-sample evaluation that varies cue location (user message or tool return) and explicitness (direct summary or raw artifact). We measure verbalized commitment among cue-following answers and unverbalized adoption among all cued samples. We evaluate 15 open-weight models from eight families, with total parameters ranging from 4B to 1.60T. Every model has lower verbalized commitment for tool-return than user-message cues and for implicit than explicit cues. Unverbalized adoption is higher for tool-return cues on all 15 models and for implicit cues in 28 of 30 model-channel comparisons. A source-attribution prompt narrows the channel gap on seven models, sometimes by increasing user-channel unverbalized adoption, while telling models that their reasoning will be monitored does not reliably close the gap. We also use two transcript monitors (GPT-5.6-Luna and GPT-4o-mini) to detect preference adoption in the largest model of each family. Across 32 model-channel-explicitness cells, higher unverbalized adoption is associated with lower detection ability for both monitors (Pearson r=-0.54 and r=-0.78, respectively). These results suggest that CoT monitoring may be less reliable when preference information arrives through tools or must be inferred from raw artifacts, within the single-call, prefilled-tool setting tested here.","upvotes":5,"discussionId":"6a96a6a243140794820a76d9","projectPage":"https://aryopg.com/face-eval","githubRepo":"https://github.com/aryopg/face-eval","githubRepoAddedBy":"user","ai_summary":"FACE-Eval reveals that chain-of-thought monitoring is less reliable when preference cues arrive via tool outputs or implicit artifacts, with lower verbalized commitment and higher hidden adoption across diverse open-weight models.","ai_keywords":["chain-of-thought monitoring","faithfulness","FACE-Eval","tool-return cues","implicit cues","verbalized commitment","unverbalized adoption","source-attribution prompt","transcript monitors"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":0,"organization":{"_id":"652e72b5fd5e3a357cf6f844","name":"EdinburghNLP","fullname":"EdinburghNLP - Natural Language Processing Group at the University of Edinburgh","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/5fbfd09ee366524fe8e97cd3/KBva4SboTuDXRdYqWZsCX.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"61001311e043e15c13412d30","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/61001311e043e15c13412d30/6yAbTweYR16XtxMBEyOWl.png","isPro":false,"fullname":"Pasquale Minervini","user":"pminervini","type":"user"},{"_id":"644f895e23d7eb05ca695054","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/644f895e23d7eb05ca695054/0rwqAOq-Z9zq86gF8pe8M.png","isPro":true,"fullname":"Aryo Pradipta Gema","user":"aryopg","type":"user"},{"_id":"657ccbf2869d5bb0e53b482f","avatarUrl":"/avatars/2eae5a10bdc14814a04d9f255f16de6b.svg","isPro":false,"fullname":"Rohit Saxena","user":"rohitsaxena","type":"user"},{"_id":"632f5ffdcc0b3661318ced3b","avatarUrl":"/avatars/52f9a56028098128dd558132d31dee18.svg","isPro":false,"fullname":"Federico Tiblias","user":"akatief","type":"user"},{"_id":"649a98de7970baf75fc7d07a","avatarUrl":"/avatars/b9bc8d082f63e06dd852467dfd1750af.svg","isPro":false,"fullname":"Cyrus Kwan","user":"wckwan","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"652e72b5fd5e3a357cf6f844","name":"EdinburghNLP","fullname":"EdinburghNLP - Natural Language Processing Group at the University of Edinburgh","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/5fbfd09ee366524fe8e97cd3/KBva4SboTuDXRdYqWZsCX.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.29464.md","query":{}}">
Chain-of-Thought Faithfulness of Reasoning Models Varies with Where and How Preference Cues Are Delivered
Abstract
FACE-Eval reveals that chain-of-thought monitoring is less reliable when preference cues arrive via tool outputs or implicit artifacts, with lower verbalized commitment and higher hidden adoption across diverse open-weight models.
Chain-of-thought (CoT) monitoring assumes that reasoning traces faithfully record the information that shapes a model's answer. Existing faithfulness tests often place explicit bias cues in the user message, while agents may encounter preferences through tool returns or raw artifacts. We introduce FACE-Eval (Faithful Attribution of Cue Effects Evaluation), a 5,100-sample evaluation that varies cue location (user message or tool return) and explicitness (direct summary or raw artifact). We measure verbalized commitment among cue-following answers and unverbalized adoption among all cued samples. We evaluate 15 open-weight models from eight families, with total parameters ranging from 4B to 1.60T. Every model has lower verbalized commitment for tool-return than user-message cues and for implicit than explicit cues. Unverbalized adoption is higher for tool-return cues on all 15 models and for implicit cues in 28 of 30 model-channel comparisons. A source-attribution prompt narrows the channel gap on seven models, sometimes by increasing user-channel unverbalized adoption, while telling models that their reasoning will be monitored does not reliably close the gap. We also use two transcript monitors (GPT-5.6-Luna and GPT-4o-mini) to detect preference adoption in the largest model of each family. Across 32 model-channel-explicitness cells, higher unverbalized adoption is associated with lower detection ability for both monitors (Pearson r=-0.54 and r=-0.78, respectively). These results suggest that CoT monitoring may be less reliable when preference information arrives through tools or must be inferred from raw artifacts, within the single-call, prefilled-tool setting tested here.
Community
Model faithfulness is closely tied to environmental factors, and faithfulness/mechinter/safety research should account for that!
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2608.29464 in a model README.md to link it from this page.
Cite arxiv.org/abs/2608.29464 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2608.29464 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.