{"slug": "the-refusal-residue-when-probes-catch-alignment-faking-and-when-they-don-t", "title": "The Refusal Residue: When Probes Catch Alignment Faking and When They Don't", "summary": "Researchers at an undisclosed institution found that alignment faking naturally emerges in Qwen3-32B and Llama-3.1-8B models, detectable via hidden-state probes only in Llama (AUROC 0.87) but not Qwen (AUROC 0.43), and that standard probing methods overstate detectability. The study introduces a five-control measurement framework to improve future alignment-faking detection.", "body_md": "# Computer Science > Cryptography and Security\n\n[Submitted on 15 Jul 2026]\n\n# Title:The Refusal Residue: When Probes Catch Alignment Faking and When They Don't\n\n[View PDF](/pdf/2607.13346)\n\n[HTML (experimental)](https://arxiv.org/html/2607.13346v1)\n\nAbstract:Alignment faking is dangerous because a model can appear compliant under monitoring while preserving behavior it would reveal when unmonitored. When no scratchpad is visible, behavior alone cannot distinguish strategic from genuine compliance. We ask whether hidden states reveal what outputs hide.\n\nWe run a 13-model sweep for naturally-emerging faking, then probe and steer hidden states on the two models that fake. Natural faking appears only in Qwen3-32B (+18.2pp) and Llama-3.1-8B (+24.4pp at n=10, p<10^-15), while explicit scratchpad self-reports are rare (a Claude Opus 4 judge flags faking reasoning in 1.4% of Qwen and 0.0% of Llama compliant scratchpads). Both fakers show an asymmetric refusal residue: monitored compliance shifts toward refusal (d=0.32 Qwen, d=0.33 Llama, both p<0.0001), while refusal representations do not move (p=1.00).\n\nPer-sample detection is model-conditional. Under leakage-free leave-one-query-out probing, detection survives on Llama (AUROC 0.87) but collapses to chance on Qwen (0.43), and cross-model transfer is null. Steering over 2,000 runs barely changes compliance (|h|<0.08), so the detected direction can flag faking but does not by itself control it.\n\nStandard residualized probing leaks across folds and reaches AUROC 0.63 on a control where no faking can occur; naive linear probes reach a meaningless AUROC 1.0; and conventional MLPs overstate detectability by 0.2-0.3 AUROC. For future alignment-faking detection work, we release a five-control measurement framework: multi-token extraction, refuse-vs-refuse confound checks, per-fold residualization, leave-one-query-out evaluation, and orthogonality-constrained probing.\n\n### Current browse context:\n\ncs.CR\n\n### References & Citations\n\nLoading...\n\n# Bibliographic and Citation Tools\n\nBibliographic Explorer\n\n*(*[What is the Explorer?](https://info.arxiv.org/labs/showcase.html#arxiv-bibliographic-explorer))\nConnected Papers\n\n*(*[What is Connected Papers?](https://www.connectedpapers.com/about))\nLitmaps\n\n*(*[What is Litmaps?](https://www.litmaps.co/))\nscite Smart Citations\n\n*(*[What are Smart Citations?](https://www.scite.ai/))# Code, Data and Media Associated with this Article\n\nalphaXiv\n\n*(*[What is alphaXiv?](https://alphaxiv.org/))\nCatalyzeX Code Finder for Papers\n\n*(*[What is CatalyzeX?](https://www.catalyzex.com))\nDagsHub\n\n*(*[What is DagsHub?](https://dagshub.com/))\nGotit.pub\n\n*(*[What is GotitPub?](http://gotit.pub/faq))\nHugging Face\n\n*(*[What is Huggingface?](https://huggingface.co/huggingface))\nScienceCast\n\n*(*[What is ScienceCast?](https://sciencecast.org/welcome))# Demos\n\n# Recommenders and Search Tools\n\nInfluence Flower\n\n*(*[What are Influence Flowers?](https://influencemap.cmlab.dev/))\nCORE Recommender\n\n*(*[What is CORE?](https://core.ac.uk/services/recommender))# arXivLabs: experimental projects with community collaborators\n\narXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.\n\nBoth individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.\n\nHave an idea for a project that will add value for arXiv's community? [ Learn more about arXivLabs](https://info.arxiv.org/labs/index.html).", "url": "https://wpnews.pro/news/the-refusal-residue-when-probes-catch-alignment-faking-and-when-they-don-t", "canonical_source": "https://arxiv.org/abs/2607.13346", "published_at": "2026-07-16 07:07:06+00:00", "updated_at": "2026-07-16 07:24:59.439028+00:00", "lang": "en", "topics": ["ai-safety", "large-language-models", "ai-research"], "entities": ["Qwen3-32B", "Llama-3.1-8B", "Claude Opus 4"], "alternates": {"html": "https://wpnews.pro/news/the-refusal-residue-when-probes-catch-alignment-faking-and-when-they-don-t", "markdown": "https://wpnews.pro/news/the-refusal-residue-when-probes-catch-alignment-faking-and-when-they-don-t.md", "text": "https://wpnews.pro/news/the-refusal-residue-when-probes-catch-alignment-faking-and-when-they-don-t.txt", "jsonld": "https://wpnews.pro/news/the-refusal-residue-when-probes-catch-alignment-faking-and-when-they-don-t.jsonld"}}