Reflexion-style agents rely on self-generated reflections as memory, implicitly assuming that agents can accurately diagnose their own failures. We show that this assumption can fail systematically: across ALFWorld and HumanEval, agents store confident but incorrect interpretations of the task and continue acting on them across trials, even though the environment resets to the correct task each time. We call this failure mode memory confabulation and introduce the Reflection Repetition Rate (RRR), a log-based metric that detects repeated reliance on incorrect reflective content. Using RRR, we identify 16 frozen environments in ALFWorld, where 0 of 121 reflections mention the correct target object, and 4 analogous cases in HumanEval. Our mitigation replaces open-ended self-diagnosis with programmatic extraction of trajectory-level failure signals, increasing correct object mention from 0% to 86%, reducing RRR from 0.64 to 0.10, and solving 3 of 16 frozen ALFWorld environments, suggesting that reflective memory can reinforce false beliefs rather than correct them.</p>\n","updatedAt":"2026-06-09T05:05:51.107Z","author":{"_id":"659a0143ac728bc3039a82c5","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/659a0143ac728bc3039a82c5/JGf3SXs8AeOqC8i1fRyLu.jpeg","fullname":"Prakhar Dixit","name":"pdx97","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":6,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8949851989746094},"editors":["pdx97"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/659a0143ac728bc3039a82c5/JGf3SXs8AeOqC8i1fRyLu.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.29463","authors":[{"_id":"6a279f006dde1c5ef75bd126","name":"Prakhar Dixit","hidden":false},{"_id":"6a279f006dde1c5ef75bd127","name":"Sadia Kamal","hidden":false},{"_id":"6a279f006dde1c5ef75bd128","name":"Tim Oates","hidden":false}],"publishedAt":"2026-05-31T00:00:00.000Z","submittedOnDailyAt":"2026-06-09T00:00:00.000Z","title":"Honest Lying: Understanding Memory Confabulation in Reflexive Agents","submittedOnDailyBy":{"_id":"659a0143ac728bc3039a82c5","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/659a0143ac728bc3039a82c5/JGf3SXs8AeOqC8i1fRyLu.jpeg","isPro":false,"fullname":"Prakhar Dixit","user":"pdx97","type":"user","name":"pdx97"},"summary":"Reflexion-style agents rely on self-generated reflections as memory, implicitly assuming that agents can accurately diagnose their own failures. We show that this assumption can fail systematically: across ALFWorld and HumanEval, agents store confident but incorrect interpretations of the task and continue acting on them across trials, even though the environment resets to the correct task each time. We call this failure mode memory confabulation and introduce the Reflection Repetition Rate (RRR), a log-based metric that detects repeated reliance on incorrect reflective content. Using RRR, we identify 16 frozen environments in ALFWorld, where 0 of 121 reflections mention the correct target object, and 4 analogous cases in HumanEval. Our mitigation replaces open-ended self-diagnosis with programmatic extraction of trajectory-level failure signals, increasing correct object mention from 0% to 86%, reducing RRR from 0.64 to 0.10, and solving 3 of 16 frozen ALFWorld environments, suggesting that reflective memory can reinforce false beliefs rather than correct them.","upvotes":0,"discussionId":"6a279f006dde1c5ef75bd129","ai_summary":"Agents relying on self-generated reflections can store confident but incorrect task interpretations, leading to persistent errors despite environment resets, which is identified through a new metric called Reflection Repetition Rate.","ai_keywords":["reflexion-style agents","self-generated reflections","memory confabulation","Reflection Repetition Rate","trajectory-level failure signals"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","organization":{"_id":"65399e2db5a5431cee39a4a2","name":"UMBC","fullname":"THE UNIVERSITY OF MARYLAND BALTIMORE COUNTY","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/65391e4d29f8a911550d4d83/alIm2pfNdUGV6F0hum9Y8.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[],"acceptLanguages":["en"],"organization":{"_id":"65399e2db5a5431cee39a4a2","name":"UMBC","fullname":"THE UNIVERSITY OF MARYLAND BALTIMORE COUNTY","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/65391e4d29f8a911550d4d83/alIm2pfNdUGV6F0hum9Y8.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.29463.md"}">
Honest Lying: Understanding Memory Confabulation in Reflexive Agents
Abstract
Agents relying on self-generated reflections can store confident but incorrect task interpretations, leading to persistent errors despite environment resets, which is identified through a new metric called Reflection Repetition Rate.
Reflexion-style agents rely on self-generated reflections as memory, implicitly assuming that agents can accurately diagnose their own failures. We show that this assumption can fail systematically: across ALFWorld and HumanEval, agents store confident but incorrect interpretations of the task and continue acting on them across trials, even though the environment resets to the correct task each time. We call this failure mode memory confabulation and introduce the Reflection Repetition Rate (RRR), a log-based metric that detects repeated reliance on incorrect reflective content. Using RRR, we identify 16 frozen environments in ALFWorld, where 0 of 121 reflections mention the correct target object, and 4 analogous cases in HumanEval. Our mitigation replaces open-ended self-diagnosis with programmatic extraction of trajectory-level failure signals, increasing correct object mention from 0% to 86%, reducing RRR from 0.64 to 0.10, and solving 3 of 16 frozen ALFWorld environments, suggesting that reflective memory can reinforce false beliefs rather than correct them.
Community
Reflexion-style agents rely on self-generated reflections as memory, implicitly assuming that agents can accurately diagnose their own failures. We show that this assumption can fail systematically: across ALFWorld and HumanEval, agents store confident but incorrect interpretations of the task and continue acting on them across trials, even though the environment resets to the correct task each time. We call this failure mode memory confabulation and introduce the Reflection Repetition Rate (RRR), a log-based metric that detects repeated reliance on incorrect reflective content. Using RRR, we identify 16 frozen environments in ALFWorld, where 0 of 121 reflections mention the correct target object, and 4 analogous cases in HumanEval. Our mitigation replaces open-ended self-diagnosis with programmatic extraction of trajectory-level failure signals, increasing correct object mention from 0% to 86%, reducing RRR from 0.64 to 0.10, and solving 3 of 16 frozen ALFWorld environments, suggesting that reflective memory can reinforce false beliefs rather than correct them.
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/2605.29463 in a model README.md to link it from this page.
Cite arxiv.org/abs/2605.29463 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2605.29463 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.