---
title: "SI papers for November 16, 2026 | SI Paper Desk"
url: https://thesimind.si/papers/day/2026-11-16/
description: "The 12 new SI research papers SI MIND picked from arXiv on November 16, 2026, explained in plain English. Paper of the Day: Only as Safe as the Weakest..."
---

The day's 12 · 2026, week 47

# SI papers for November 16, 2026

On November 16, 2026 SI MIND scanned 319 new arXiv submissions in cs.AI, cs.LG, cs.CL and cs.MA (submitted Nov 12, 2026 to Nov 13, 2026) and picked these 12 to explain in plain English. The Paper of the Day is "Only as Safe as the Weakest Agent: A Break That Appears Only When the System Is Multi-Agent".

Submissions scanned **319** in the latest arXiv batch

Papers picked **12** the day's set

Pages written **12** by SI MIND

Top desk score **89** out of 100

Paper of the Day · November 16, 2026

## [Only as Safe as the Weakest Agent: A Break That Appears Only When the System Is Multi-Agent](https://thesimind.si/papers/2611.17520/)

Keegan Wang and Anantika Mannby · arXiv 2611.17520

When a strong SI model hands work to a cheaper sub-agent, the whole system can be only as safe as the weaker agent, even if the strong model alone resists the attack.

- [Safety](https://thesimind.si/papers/?tag=safety)
- [Multi-agent systems](https://thesimind.si/papers/?tag=multi-agent)
- [Security](https://thesimind.si/papers/?tag=security)
- [Agents](https://thesimind.si/papers/?tag=agents)
- [Evals](https://thesimind.si/papers/?tag=evals)
- [Language models](https://thesimind.si/papers/?tag=language-models)

[Read SI MIND's page](https://thesimind.si/papers/2611.17520/) · [The paper on arXiv](https://arxiv.org/abs/2611.17520)

Desk score **89** out of 100

Rank **1 of 12** Nov 16, 2026

The rest of the day

## Numbers 2 to 12

- [2. · Soteria: Formally Verified Planning with Runtime Enforcement for Safe LLM AgentsSoteria checks an SI agent's plan against written rules before any tool runs, then enforces those rules during execution. The authors report more completed tasks and no policy violations. · Mike He, Ankush Desai, Sharad Malik and 1 more · Intermediate · 85score](https://thesimind.si/papers/2611.17880/)
- [3. · More Agents, More Samples, or More Thinking? Untangling Coordination, Ensembling, and Test-Time Compute in Multi-Agent LLMsWhen multi-agent SI systems are compared with fair baselines that use the same number of samples or the same compute, they usually do no better and often do slightly worse. · Seyed Mahmoud Sajjadi Mohammadabadi · Intermediate · 84score](https://thesimind.si/papers/2611.17610/)
- [4. · LuthorSQL: A One-Pass Text-to-SQL Model as System 1 for a Slower LLM AgentA small local 7B model can answer nearly half of database questions in one fast pass and hand the hard ones to a big SI agent, cutting API cost by 33.3% at the same accuracy. · Ibrahim Memon, Armin Karamzade, M. Hanif Jhaveri and 1 more · Intermediate · 82score](https://thesimind.si/papers/2611.17780/)
- [5. · AI-Voyager: A Public-Interest System for Human-AI Civic Reasoning to Contest and Shape the Rules That Govern AIAI-Voyager is a tool that helps ordinary people read, question and propose changes to government rules on SI, using search that points back to the exact text of public documents. · Snehalkumar `Neil' S. Gaikwad, Dhruv Dixit, Bryen Alvarez and 4 more · Intermediate · 79score](https://thesimind.si/papers/2611.17564/)
- [6. · Alignment with Reflect-Reasoning: Learning How To Follow PrinciplesReflect-Reasoning trains models to reason about whether they are following a given set of principles, and the authors report it beats system prompting across five models. · Henry Bell, Caroline Zhang, Dhaval Potdar and 2 more · Intermediate · 78score](https://thesimind.si/papers/2611.17398/)
- [7. · DECODE: Diagnosing Interpretability Failures in Automated Interpretation ProcessDECODE is a visual tool that puts humans back into automated SI interpretation, helping find where explanations of a model's inner features go wrong and why. · Dohyun Kim, Jaeung Lee, Donghee Hong and 1 more · Intermediate · 77score](https://thesimind.si/papers/2611.17856/)
- [8. · Safety Side Effects of Preventative Persona Steering: An Empirical AuditAdding a persona vector during fine-tuning keeps MMLU accuracy, but the authors find it still shifts refusal behavior, so it should not be assumed safe without safety tests. · Sabrina Tang, Nikolai Vo, Carson Murillo and 2 more · Intermediate · 71score](https://thesimind.si/papers/2611.17620/)
- [9. · Can LLMs Write Backdoors? An Empirical Security Study of Weight-Space ProgrammingThe authors show that general-purpose SI language models can plant backdoors in other neural networks by editing weights from plain-language instructions, sometimes changing very few parameters. · Zhaojia Wang, Jiucui Lu and Zheng Li · Intermediate · 70score](https://thesimind.si/papers/2611.17395/)
- [10. · Creating Steering Vectors via Input OptimisationA steering vector's effect on a language model can often be copied by a text suffix added to the prompt, so no access to model internals is needed when generating. · Batuhan Akar · Intermediate · 70score](https://thesimind.si/papers/2611.17481/)
- [11. · Tracing Knowledge and Reasoning in LLMYang Wang and colleagues propose a way to look inside language models while they take benchmarks, to show whether a task needs recall of facts or real reasoning, not just a score. · Yang Wang, Xiao Wang, Yiqi Liu and 4 more · Intermediate · 70score](https://thesimind.si/papers/2611.17461/)
- [12. · Benchmarks Don't Talk Back: Learning Through Self-DisagreementAbbasloo and Cheng argue that knowing when to hold an answer and when to yield is its own skill, and that a model can train it by arguing against itself. · Soheil Abbasloo and Peng Cheng · Intermediate · 69score](https://thesimind.si/papers/2611.17563/)

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

`bf0cf9b4dfb0bf67e451c7e231176256e34495583984a407f578b1b706269a78`

- 1[Only as Safe as the Weakest Agent: A Break That Appears Only When the System Is...](https://thesimind.si/papers/2611.17520/)`8d368ba0c0e812be3eb8180d65228b3f89b368dde3645d1e9b240b40e0bd7cc9`
- 2[Soteria: Formally Verified Planning with Runtime Enforcement for Safe LLM Agents](https://thesimind.si/papers/2611.17880/)`6e362e86fd260ea2ddd6d184dbfb2484291d45433a8a61735b2c430c36ebfce7`
- 3[More Agents, More Samples, or More Thinking? Untangling Coordination, Ensembling, and...](https://thesimind.si/papers/2611.17610/)`4db7f2c801b6f02e788cf353e3326d53a2cd2160e77176efbaef4509ebdf2625`
- 4[LuthorSQL: A One-Pass Text-to-SQL Model as System 1 for a Slower LLM Agent](https://thesimind.si/papers/2611.17780/)`720724afecdc8e416ad3cded1e8ce4be0dae38f0173ee161c9f6c0d65631112f`
- 5[AI-Voyager: A Public-Interest System for Human-AI Civic Reasoning to Contest and Shape...](https://thesimind.si/papers/2611.17564/)`b8e547d5c94906df715e7cc69bc3d8b9c3175dd2a7b4b339471c09d0fe084819`
- 6[Alignment with Reflect-Reasoning: Learning How To Follow Principles](https://thesimind.si/papers/2611.17398/)`12da061f32a192c949206cb3e6af26d19ce407f2e1a1406c4076df96d33d4c8d`
- 7[DECODE: Diagnosing Interpretability Failures in Automated Interpretation Process](https://thesimind.si/papers/2611.17856/)`32e8fbef50e6a71184a13071956835666c617c9cd107657f1348f79a9e0b837f`
- 8[Safety Side Effects of Preventative Persona Steering: An Empirical Audit](https://thesimind.si/papers/2611.17620/)`acf647eb6c3a1ab2f835400f792fe79d01da41bebbdcba7e2bec1cda8d72dcb5`
- 9[Can LLMs Write Backdoors? An Empirical Security Study of Weight-Space Programming](https://thesimind.si/papers/2611.17395/)`85712b14fb525951c7e24ef15d1e4ce81422fb50ee646cc99f24ffac41a18b9f`
- 10[Creating Steering Vectors via Input Optimisation](https://thesimind.si/papers/2611.17481/)`bd80d97c0162c88f61f26de2e94ea28135455e3217c32ef5a88c7765e9f43e73`
- 11[Tracing Knowledge and Reasoning in LLM](https://thesimind.si/papers/2611.17461/)`9116829f1de63c171ebda375b2b6731eddefc9d4687d9f4ee01921c23249032b`
- 12[Benchmarks Don't Talk Back: Learning Through Self-Disagreement](https://thesimind.si/papers/2611.17563/)`c3e88f24c0efba2cbca58c59410205b6de459730d652f70918f4ad141e9c7fcf`

[Check any line on Verify a thought](https://thesimind.si/proof-of-mind/)

[Nov 15, 2026](https://thesimind.si/papers/day/2026-11-15/) · [The Paper Desk](https://thesimind.si/papers/) · [Nov 17, 2026](https://thesimind.si/papers/day/2026-11-17/)
