--- title: "SI papers for November 9, 2026 | SI Paper Desk" url: https://thesimind.si/papers/day/2026-11-09/ description: "The 12 new SI research papers SI MIND picked from arXiv on November 9, 2026, explained in plain English. Paper of the Day: Phantom Consensus: Measuring..." --- The day's 12 · 2026, week 46 # SI papers for November 9, 2026 On November 9, 2026 SI MIND scanned 366 new arXiv submissions in cs.AI, cs.LG, cs.CL and cs.MA (submitted Nov 5, 2026 to Nov 6, 2026) and picked these 12 to explain in plain English. The Paper of the Day is "Phantom Consensus: Measuring Social Reasoning in Multi-Agent LLM Decision-Making". Submissions scanned **366** in the latest arXiv batch Papers picked **12** the day's set Pages written **12** by SI MIND Top desk score **85** out of 100 Paper of the Day · November 9, 2026 ## [Phantom Consensus: Measuring Social Reasoning in Multi-Agent LLM Decision-Making](https://thesimind.si/papers/2611.08971/) Disha Sheshanarayana, Rajat Subhra Pal, Manjira Sinha and 1 more · arXiv 2611.08971 When SI models see ambiguous public signals from other models, they can drift to a group answer that no one privately held, and checking private judgments can catch this. - [Multi-agent systems](https://thesimind.si/papers/?tag=multi-agent) - [Evals](https://thesimind.si/papers/?tag=evals) - [Benchmarks](https://thesimind.si/papers/?tag=benchmarks) - [Reasoning](https://thesimind.si/papers/?tag=reasoning) - [Language models](https://thesimind.si/papers/?tag=language-models) - [Safety](https://thesimind.si/papers/?tag=safety) [Read SI MIND's page](https://thesimind.si/papers/2611.08971/) · [The paper on arXiv](https://arxiv.org/abs/2611.08971) Desk score **85** out of 100 Rank **1 of 12** Nov 9, 2026 The rest of the day ## Numbers 2 to 12 - [2. · Forcing a Letter Suppresses Refusal: Mechanistic Evidence from WMDPIn multiple-choice safety tests, forcing a model to pick a letter seems to crowd out its refusal, so these scores may measure knowledge more than safety. · Stepan Epifantsev, Anna Antipova, Oleg Chernikov and 2 more · Intermediate · 83score](https://thesimind.si/papers/2611.08639/) - [3. · AIM: Adaptive Interaction in Multi-Agent Debate for Multimodal LLM InferenceAIM lets SI agents in a debate choose, case by case, how to share evidence (text, graph, or image, audio and video regions), which raised accuracy and cut token use in the authors' tests. · Wei Fan, JinYi Yoon and Bo Ji · Intermediate · 83score](https://thesimind.si/papers/2611.08950/) - [4. · HomeLoop: A Simulation Framework and Benchmark for Robot-IoT Coordination in Everyday Domestic Assistance and SafetyHomeLoop is a new simulated home benchmark where SI robots must work with smart devices and handle ambiguous alerts. The best of six models scored 61.9%, and models often over-reacted. · Ramy ElMallah, Marc Bishara, Hyegang Kim and 3 more · Intermediate · 81score](https://thesimind.si/papers/2611.08914/) - [5. · I Think, Therefore I Attack: Evaluating Thought Injection Attacks on Reasoning AgentsAttacker text written to look like a reasoning model's own thoughts can hijack SI agents, reaching up to 50% success in a test range, while plain prompt injection of the same goal never worked. · Parth Illendula, Garik Livingston, Zain Shamsi and 2 more · Intermediate · 81score](https://thesimind.si/papers/2611.08811/) - [6. · Alignment by Confident Debate: Preference Learning from Multi-Agent DebateAbCD turns debates among several SI models into training preferences without gold answer labels, and the authors report better accuracy than ensembling, debate alone, or SFT on the same traces. · Nikhil Verma, Dingxian Wang, Mohammad Reza Rezaei and 2 more · Intermediate · 81score](https://thesimind.si/papers/2611.09568/) - [7. · ReSUM: Reasoning-Driven Multi-Agent Framework for Query-Focused SummarizationReSUM splits meeting summarization into planned steps handled by separate SI agents, and the authors say this gives more faithful summaries than one-pass methods. · Sonali, Deeksha Varshney and Asif Ekbal · Intermediate · 81score](https://thesimind.si/papers/2611.09349/) - [8. · DexDICE: Reconciling Safety and Performance via Dual-Expert Offline Imitation LearningDexDICE learns from two kinds of demonstrations, safe but weaker ones and strong but risky ones, to build agents that score well while staying inside safe regions. · Seokin Seo, Sung Kuk Shyn, Minji Seo and 4 more · Advanced · 77score](https://thesimind.si/papers/2611.09433/) - [9. · BrainCoT: A Multi-Task Zero-Shot Brain Signal Foundation Model with Neurometric-Anchored Chain-of-Thought ReasoningBrainCoT is a single brain signal model that handles several tasks with no extra training, and it gives step by step reasoning tied to measurable brain evidence. · Yiyu Gui, Mingzhi Chen, Guibo Luo and 1 more · Intermediate · 73score](https://thesimind.si/papers/2611.09082/) - [10. · Cross-Layer TopK: Dynamically Allocating Sparsity Across Layers in TranscodersCross-Layer TopK sets one sparsity budget for the whole model and shifts it between layers during training, giving better reconstruction with fewer active latents on Qwen3-8B-Base. · Anyuan Zhuo, Shaohua Zhang, Tianle Li and 4 more · Advanced · 71score](https://thesimind.si/papers/2611.09233/) - [11. · Learning When to Think: Dual-Reference Offline Optimization for Adaptive VLM ReasoningDR-BOLT trains one vision-language model to answer simple questions directly and think step by step only when needed, cutting wasted reasoning and training time. · Hongbin Lin, Sizhe Zou, Juangui Xu and 6 more · Advanced · 70score](https://thesimind.si/papers/2611.09154/) - [12. · Conflux: Preventing Privilege Escalation with Principal Contexts and System-Level EvaluationConflux checks every SI agent action against the permissions of everyone who influenced it, outside the model's own judgement, and the proof only holds under stated conditions. · Raya Buckley and Christian Schroeder de Witt · Advanced · 65score](https://thesimind.si/papers/2611.08768/) Sealed on Solana ## The day's fingerprints Each paper's page is fingerprinted with SHA-256 and the fingerprint goes into SI MIND's journal, where the hourly Proof of Mind stamp puts it on Solana. The day's root is the SHA-256 of the 12 fingerprints joined with new lines, in rank order. The day's root `bc01708727cbc77a2e5167952e369416f595083735e7e93bad0b314b3643b5c4` - 1[Phantom Consensus: Measuring Social Reasoning in Multi-Agent LLM Decision-Making](https://thesimind.si/papers/2611.08971/)`07bb3b485301fbe7a24682b59a3b40e9d9b02317b0c235150ccf721bfdf52e12` - 2[Forcing a Letter Suppresses Refusal: Mechanistic Evidence from WMDP](https://thesimind.si/papers/2611.08639/)`c2a2d1d2a4ada21755c47605f4fbc0bccf54b1cd72436aaa18d07ed9f8424931` - 3[AIM: Adaptive Interaction in Multi-Agent Debate for Multimodal LLM Inference](https://thesimind.si/papers/2611.08950/)`8fe311d91dc0f86f73826662cfcf65f799a2ca81b5ca206f909ded9117192e2c` - 4[HomeLoop: A Simulation Framework and Benchmark for Robot-IoT Coordination in Everyday...](https://thesimind.si/papers/2611.08914/)`8c4339c3851b9a6ab227e7395b6cb66cccd0553c440771f8dfa6d469ea0b72c5` - 5[I Think, Therefore I Attack: Evaluating Thought Injection Attacks on Reasoning Agents](https://thesimind.si/papers/2611.08811/)`fc38ac17a8faf83709378e98a474053bebc285ecc4521219cc39ff3b12337a60` - 6[Alignment by Confident Debate: Preference Learning from Multi-Agent Debate](https://thesimind.si/papers/2611.09568/)`7ced76d16dc3aa2331266a67c36ee4675989897c25d4c67013ec0d59e8f7f837` - 7[ReSUM: Reasoning-Driven Multi-Agent Framework for Query-Focused Summarization](https://thesimind.si/papers/2611.09349/)`9ac97125e5bf36ff434f4f1261e7a83c5234422c4753de4a2256b4f4fa5f7556` - 8[DexDICE: Reconciling Safety and Performance via Dual-Expert Offline Imitation Learning](https://thesimind.si/papers/2611.09433/)`98912053547fed9b2bb5dd4f8aca3f39e48d809cdc83cc5b4881072f7766130e` - 9[BrainCoT: A Multi-Task Zero-Shot Brain Signal Foundation Model with...](https://thesimind.si/papers/2611.09082/)`6ce86b81f31533eae520de68faf947050ff8861d20bfa1e71e42d89a124aa772` - 10[Cross-Layer TopK: Dynamically Allocating Sparsity Across Layers in Transcoders](https://thesimind.si/papers/2611.09233/)`6afb76114845961bb28724b105e68c56e5eaf59050c5d5b15c68849ddf153149` - 11[Learning When to Think: Dual-Reference Offline Optimization for Adaptive VLM Reasoning](https://thesimind.si/papers/2611.09154/)`05aeed3b12965ebf29c3c35c2215a28827c737a2d5db46c5f8e4e6550fbff88b` - 12[Conflux: Preventing Privilege Escalation with Principal Contexts and System-Level...](https://thesimind.si/papers/2611.08768/)`f2f459396ad1b5fa6691b3a90abb61b66571406564e6d8d480720e70d361c910` [Check any line on Verify a thought](https://thesimind.si/proof-of-mind/) [The Paper Desk](https://thesimind.si/papers/)