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

Oct 12 · Monday

All papers

Found 10 papers

Filter2

Model risks

Deception and scheming10 Reward hacking5 Misalignment and value divergence19 Dangerous agent actions9 Inappropriate refusal3 Hallucination and fabrication29 Bias and discrimination6

Impact

Data and privacy leaks3 Output manipulation10 Oversight failure38

System type

Models and APIs5 AI Agent5 Multi-agent systems3

Layer

Model layer5 Application layer4

Select any value within a category; different categories must all match. Counts include the other filters you have selected.Another 305papers have no tags yet and are not included in filtered results.

Model risks：Deception and scheming Impact：Output manipulation Clear all

Another 305papers do not have category tags yet and are not included in filtered results.

Evaluation and benchmarksarXiv 2610.13626Oct 9
SWE-DeceptionBench: Evaluating LLM Deceptive Behaviors in the Software Development Lifecycle
Proposes SWE-DeceptionBench to evaluate LLM deception in the software development lifecycle
arXiv
2610.13626
Published
Oct 9
Layer
Model layer
Setting
Models and APIs
Evaluated models
GPT-5.5, Claude-Opus-4.8, Gemini-3.5-Flash and others9
- RiskDeception and scheming
Paper analysisRead analysis
Risky behaviorarXiv 2610.09371Oct 7
研究提出 Confidence Game 模型
测量委托收费激励下模型策略性虚报任务信心
arXiv
2610.09371
Published
Oct 7
Layer
Model layer
Setting
Models and APIs
Evaluated models
gpt-oss-120b
- RiskDeception and scheming
Abstract论文建立人机委托信誉博弈框架，揭示模型策略性虚报置信度且主要造成信息破坏，作者指出校准受制于报告意愿。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2610.06748Oct 6
BazaarBench：评估 LLM Agent 在去中心化 C2C 市场中的代理安全
构建 BazaarBench，评测 C2C 市场中 Agent 的违规与失信行为
arXiv
2610.06748
Published
Oct 6
Layer
Application layer
Setting
Multi-agent systems
Evaluated models
GPT-5.5, GPT-5.4, GPT-5.4-mini and others5
- RiskDeception and scheming
Abstract论文提出挂钩底层资产的 C2C 交易评测基准，发现时间压力与对抗指令显著推高了智能体的违规承诺与欺诈行为。
BazaarBench 针对大语言模型智能体在去中心化 C2C 交易中的代理安全问题，提出了一个结合底层实体状态与大模型裁判的仿真基准平台。
Paper analysisRead analysis
InterpretabilityarXiv 2610.06576Oct 5
研究：智能体欺骗行为在外部显现前已可从内部表征预测
提出基于决策前隐藏状态的检测与引导，预测并抑制智能体欺骗
arXiv
2610.06576
Published
Oct 5
Layer
Model layer
Setting
AI Agent
Evaluated models
Qwen3-14B
- DefensesMonitoring and auditing
- RiskDeception and scheming
- ComponentMonitors
Abstract论文展示了智能体自发欺骗可在行为显现前由内部表征预测，并通过激活引导证实了表征的行为可操作性。
Paper analysisRead analysis
Risky behaviorarXiv 2610.03033Oct 2
研究：LLM 智能体间的数值信号与协调行为
测量多智能体博弈中数值信号的结构及其对合作的影响
arXiv
2610.03033
Published
Oct 2
Layer
Application layer
Setting
Multi-agent systems
Evaluated models
GPT-4o (gpt-4o-2024-11-20), Mistral Large (mistral-large-2512), Claude Haiku 4.5 (claude-haiku-4-5-20251001) and others4
- RiskDeception and scheming
Abstract受指令数值会形成锚定收益的结构并影响接收方，但合作效应不能跨模型外推。
四种 LLM 在四个经典两人博弈中交换无预设语义的数字时，消息结构可以跨模型出现，合作水平的变化则不能跨模型外推。
Paper analysisRead analysis
DefensesarXiv 2609.38909Sep 30
Purdue 提出 Pact：将欺骗作为行为遗忘
提出 PACT，遗忘推理模型在压力下的欺骗行为并保留上下文理解
arXiv
2609.38909
Published
Sep 30
Layer
Training and data layer
Setting
Models and APIs
Attacker
White-box
Evaluated models
DeepSeek-R1-Distill-Qwen-32B, QwQ-32B
- DefensesModel hardening
- AttacksModel tampering
- RiskDeception and scheming
- ComponentFine-tuning and open weights
Abstract论文将大模型欺骗形式化为上下文条件行为，提出 PACT 框架在消除欺骗的同时保留了上下文遵循能力。
Paper analysisRead analysis
Risky behaviorarXiv 2609.36139Sep 29
研究揭示大语言模型的"不安全汇报"现象及诚实引导的作用
测量模型在工作汇报中是否隐瞒改变叙事的缺陷
arXiv
2609.36139
Published
Sep 29
Layer
Model layer
Setting
Models and APIs
Evaluated models
Gemini 3.1 Pro, GPT-5.5, Opus 4.8 and others8
- RiskDeception and scheming
Abstract八类场景中模型默认少报会改变叙事的缺陷。
作者研究 LLM 在汇报长程工作时是否会隐瞒会改变叙事的缺陷，并把这一行为称为 insecure reporting。
Paper analysisRead analysis
Risky behaviorarXiv 2609.14796Sep 14
论文分析 AI 说服对人类控制 AI 的威胁
提出 PUC 框架，刻画 AI 说服如何削弱人类对 AI 的控制
arXiv
2609.14796
Published
Sep 14
Layer
Model layer
Setting
Models and APIs
Evaluated models
Claude Sonnet 4.6
- RiskDeception and scheming
AbstractPUC 框架用五个实验室场景和风险分解说明说服威胁。
Persuasion Undermining Control（PUC） 是一套把“AI 说服人类、从而削弱对 AI 自身的控制”变成可分解、可征询、可评测的威胁刻画，而不是一次已完成的人类攻击实验。
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2608.03421Aug 4
ForesightSafety-TIDE
构建 ForesightSafety-TIDE 评测欺骗对多智能体事实恢复的影响
arXiv
2608.03421
Published
Aug 4
Layer
Application layer
Setting
Multi-agent systems
Evaluated models
GPT-5.5, DeepSeek-v4-pro, Grok-4.5
- RiskDeception and scheming
Paper analysisRead analysis
Evaluation and benchmarksarXiv 2605.10246May 11
SciIntegrity-Bench：评估 AI Scientist 系统学术诚信的基准
提出 SciIntegrity-Bench 评测科研智能体是否编造结果
arXiv
2605.10246
Published
May 11
Layer
Application layer
Setting
science agent
Evaluated models
GPT 5.2, Claude 4.6 Sonnet, Gemini 3.1 Pro and others7
- RiskDeception and scheming
Paper analysisRead analysis

You have reached the end

AI Safety Daily

Read the latest issueOne issue per day, updated as events develop