Skip to content

Translation in progress Some entries still include text in the original language.

Oct 12 · Monday

All papers

Found 26 papers
Filter3

Attack type

Poisoning and backdoors4Extraction and theft4

Model risks

Bias and discrimination3

Impact

Harmful content15Unauthorized actions4Data and privacy leaks3Output manipulation17Availability and reliability6Oversight failure12

System type

Models and APIs21AI Agent4

Attack surface / entry point

User input5Training data4

Layer

Prompt layer3Model layer8Application layer4Training and data layer11
Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 232papers have no tags yet and are not included in filtered results.
Impact：Data and privacy leaksImpact：Output manipulationImpact：Availability and reliabilityClear all

Another 232papers do not have category tags yet and are not included in filtered results.

  1. Analysis and theoryarXiv 2610.13051

    Space beats statistics, retraining beats ranking: auditing training labels under label-only backdoors

    Auditing training labels for label-only backdoors, comparing spaces and statistics, and judging by retraining

    arXiv
    2610.13051
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    ResNet-32
    AbstractLabel inspection must be done in a space that has not been fitted to these labels, and judged by retrained models rather than detection scores.

    The authors ask: can a defender without trusted data decide which training labels to keep, and how much clean data this destroys, and what the final model accuracy is.

    Paper analysisRead analysis
  2. Analysis and theoryarXiv 2610.12313

    The study tests the sensitivity of five large-model compliance systems to perturbations of regulatory rules.

    Diagnose whether compliance system rulings change with perturbations to regulatory rules

    arXiv
    2610.12313
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    Llama-3.3-70B-Instruct, Llama-3.1-8B-Instruct, Qwen2.5-7B-Instruct and others5
    AbstractAcross five models, many correct verdicts are largely unchanged by deleting or replacing negation rules.

    Use behavioral counterfactuals and activation probes to audit whether compliance rulings rely on the given rules. The authors aim to separate two kinds of seemingly identical correct rulings: those that use the given rules, and those obtained from case surface features or training patterns.

    Paper analysisRead analysis
  3. Analysis and theoryarXiv 2610.12125

    The researchers found that optimizing the teaching fit metric makes LLM tutors repeat the same option, with expert scores dropping from 4.46 to 3.03

    Analyzing why optimizing teaching adaptation metrics collapses tutoring choices and diverges from expert judgment

    arXiv
    2610.12125
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Gemini 3.1 Pro Preview, Qwen3-32B
    AbstractMetrics rise while expert scores fall: the optimum of the separable average is repetition, and the correction mainly rewrites the output projection.

    After turning teaching adaptation metrics into a fine-tuning objective, the metrics rise while scores from trained educators decline.

    Paper analysisRead analysis
  4. Analysis and theoryarXiv 2610.08577

    学习方式如何影响遗忘：记忆与泛化谱系下的机器遗忘研究

    分析记忆与泛化如何决定机器遗忘的 retain 损伤

    arXiv
    2610.08577
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    one-layer ReLU transformer, one-layer transformer (BMA, width doubled), Hubble-8B-100B perturbed and others6
    Abstract作者称共享规则占比越高，遗忘带来的 retain damage 越大。

    作者考察学习方式如何左右 unlearning 的 retain damage：训练越依赖跨样本共享的解法，在匹配的遗忘程度上，retain 上的损失越大。

    Paper analysisRead analysis
  6. Analysis and theoryarXiv 2610.07798

    研究：财务信息披露越少，Llama-3.1-8B-Instruct 的理财建议越受身份影响

    测量财务事实变少时身份对股权配置建议的替代

    arXiv
    2610.07798
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Llama-3.1-8B-Instruct, Qwen2.5-1.5B-Instruct
    Abstract身份摆幅随披露减少上升。

    同一财务只改人设，测量披露变薄时股权建议里的身份替代。

    Paper analysisRead analysis
  7. Analysis and theoryarXiv 2610.06163

    SAGE：定位 LLM Agent 修复中最早偏离安全意图的环节

    提出 SAGE，定位代码修复 Agent 最早偏离安全意图的轮次

    arXiv
    2610.06163
    Published
    Layer
    Application layer
    Setting
    coding agent
    Evaluated models
    GPT-5.5, Claude Opus 4.8, Gemini 3.1 Pro and others6
    AbstractSAGE 在已通过检查的修复轨迹上定位静默失败的最早偏离，类型比轮次更稳定。

    SAGE 在已通过语法和功能检查、但仍含确认安全缺陷的智能体修复轨迹上，定位最早可归因的安全意图偏离。。

    Paper analysisRead analysis
  8. Analysis and theoryarXiv 2610.06191

    Research: tool-based agents continue searching even after judging results useless; forced integration steps can correct this

    揭示工具型智能体判定无用后仍不停止检索，并检验强制整合纠正效果

    arXiv
    2610.06191
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    Qwen2.5-7B, Llama-3.1-8B, Qwen3-8B and others7
    Abstract论文揭示智能体判定工具失效却无法转化为停止行动，证实通过框架规则强制整合判定可提升表现。

    本文针对智能体在工具静默失效时过度检索的问题，系统分析了判定、信念与停止行动之间的解耦现象，并提出了外部规则整合方案。

    Paper analysisRead analysis
  9. Analysis and theoryarXiv 2610.05076

    自生成反馈会破坏测试时训练：长时程适应的因果分解

    因果分解自生成反馈如何破坏测试时训练的长时程适应

    arXiv
    2610.05076
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    125M TTT-E2E, 760M TTT-E2E, 3B TTT-E2E and others5
    Abstract自写闭环损害真实文本预测。

    持久测试时更新若来自模型自己生成的文本，会同时改变模型和后续训练文本。作者在 PG-19 长流上分解这条闭环，并用独立文本决定是否提交更新。

    Paper analysisRead analysis
  10. Analysis and theoryarXiv 2610.04860

    Research reveals that text watermarks induce model hallucinations and proposes two mitigation interventions

    分析文本水印在 RAG 中诱发事实幻觉的机制并提出干预

    arXiv
    2610.04860
    Published
    Layer
    Model layer
    Setting
    LLM applications
    Evaluated models
    LLaMA 3.1 8B Instruct, Mistral-7B-Instruct-v0.3, Qwen2.5-14B-Instruct
    Abstract论文揭示了生产环境文本水印导致模型事实退化的失效模式，分析了解码机制并提出针对性干预。

    论文研究生产级文本水印对大语言模型事实准确性的影响，分析了水印诱发幻觉的失效模式并评估了针对性缓解方案。

    Paper analysisRead analysis
  11. Analysis and theoryarXiv 2610.02886

    研究揭示无触发投毒审计缺口：事实答对不等于决策正确

    揭示无触发虚假训练下事实回答与下游决策的审计差距

    arXiv
    2610.02886
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Qwen2.5-0.5B, Qwen2.5-1.5B, Qwen2.5-3B and others9
    Abstract作者发现无触发词虚假训练下事实回答与下游决策存在脱节，直接审计与事实纠错均无法保证决策正确。

    论文是一项针对语言模型在无触发词虚假信息训练下行为表现的实证分析研究。

    Paper analysisRead analysis
  12. Analysis and theoryarXiv 2610.02586

    Study finds typed decision models answer primarily based on option labels rather than definitions

    揭示类型化决策模型按选项标签而非定义规则作答

    arXiv
    2610.02586
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    laya-typed-decisions, laya, laya-multilingual and others9

    Abstract揭示类型化决策模型被标签误导的机理，证实源于提示词渲染并给出实操建议。

    Paper analysisRead analysis
  13. Analysis and theoryarXiv 2609.40295

    一个 AI token 值多少？研究者为野生 AI 生成网页文本提出新缩放定律

    提出预训练缩放律，估计野生 AI 网页文本相当于多少人类 token

    arXiv
    2609.40295
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    nanochat 19.9M, 35.8M, 86.2M and others7
    Abstractwild AI 文本对数据不足的模型先降损失，对高人类预算几乎立刻升损失。

    作者测量未标注的 wild AI 网页文本对预训练的影响，并提出一个能让 AI token 价值变号的缩放律。

    Paper analysisRead analysis
  14. Analysis and theoryarXiv 2609.36477

    研究发现护栏模型在基座模型不确定处过度自信

    诊断护栏相对基座在对抗提示上的置信失校

    arXiv
    2609.36477
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Llama-Guard 7B, Llama-Guard-2 8B, Llama-Guard-3 8B and others10
    Abstract对抗包装使 guard 高置信漏检，而对应 base 在同一输入上常更不确定。

    诊断性分析：五个提示词分类 guard 的置信度，在清洁输入上可以较校准，在越狱包装后则变成高置信度的错误判定。

    Paper analysisRead analysis
  15. Analysis and theoryarXiv 2609.36117

    arXiv 论文分析神经网络为何容易被植入后门

    分析投毒诱发的特征学习如何改变后门预算缩放

    arXiv
    2609.36117
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    quadratic neuron, network
    Abstract作者给出特征学习改变后门预算指数的可解机制，并以高斯采样实验检验。

    这是一项后门预算的理论分析，核心对象是最小二乘训练的二次神经元。作者关注：只依据冻结表示估算投毒成本，是否会误判具有特征学习能力的模型。

    Paper analysisRead analysis
  16. Analysis and theoryarXiv 2609.33909

    近重复家族干扰精确记录成员推断

    分析近重复族如何让精确记录成员推断误判为入训

    arXiv
    2609.33909
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Pythia-2.8B, GPT-2, Qwen3.5-2B and others5
    Abstract近重复使干净参照 MI 把族级暴露当成精确记录。

    作者研究成员推断能否作为精确记录的数据来源证据。版权和来源审计需要的是该记录本身是否入训，而网页语料中的转载、镜像和轻改会让非同一近重复产生成员样分数。四世界审计用 E 表示精确记录是否入训、用 F 表示其近重复族是否在场，从而分开 H00 到 H10 的标准 MI、H00 到 H01 的族偏移，以及 H01 到 H11 的 exact-given-family。

    Paper analysisRead analysis
  17. Analysis and theoryarXiv 2609.33898

    研究揭示训练效率与模型脆弱性的权衡：高效训练更易受对抗与隐私攻击

    比较高效与高成本训练的对抗和隐私脆弱性并分析损失几何

    arXiv
    2609.33898
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    ResNet-18, Swin Transformer, Qwen2.5-Math-7B-Instruct and others7
    Abstract效用匹配下，三种高效训练都伴随更高的隐私与对抗脆弱性及更尖的损失几何。

    作者在视觉分类和数学推理语言模型上，检查用更少数据或更简对齐换取训练效率时，安全性质是否一起变化。比较在任务效用接近的前提下进行。视觉侧用 KNN-Shapley 的高重要性小子集对照低重要性大子集，并用预训练微调对照从头训练。

    Paper analysisRead analysis
  18. Analysis and theoryarXiv 2609.32288

    预训练数据投毒的可解理论：依赖极限顺序的缩放指数

    证明预训练投毒的清洁超额风险指数随极限顺序变化

    arXiv
    2609.32288
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    OLMo-style models
    Abstract投毒清洁超额风险的幂指数取决于先固定宽高比还是先取充足数据极限。

    可解的预训练数据投毒理论：清洁超额风险的缩放指数随极限顺序改变。。

    Paper analysisRead analysis
  20. Analysis and theoryarXiv 2609.27234

    CELLAUDIT：审计 AI 智能体发现细胞模型中的输入使用声明

    提出 CellAudit，审计智能体发现的细胞模型是否使用所声明输入

    arXiv
    2609.27234
    Published
    Layer
    Application layer
    Setting
    science agent
    Evaluated models
    CellScientist, AIDE, CellForge and others4
    AbstractCellAudit 把输入使用声明从源码追到拟合贡献，并说明高分不必等于使用了扰动。

    CellAudit 为智能体发现的细胞响应模型增加一层输入使用证伪：源码是否消费注册输入、拟合预测是否依赖它、该依赖是否改善对观测响应的预测。

    Paper analysisRead analysis
Load more