Skip to content

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

Oct 12 · Monday

All papers

Found 418 papers
Filter2

Attack type

Jailbreaks201Prompt injection14Poisoning and backdoors8Model tampering29Extraction and theft3Detection evasion15Malicious use12

Attack technique

Gradient and representation optimization24Automated search43Multi-turn escalation20Encoding and obfuscation32Role-play and persuasion17Conversation structure manipulation9Reasoning-chain manipulation7Cross-modal payloads31Defense-targeted12Dormant triggers4

Model risks

Deception and scheming40Reward hacking51Misalignment and value divergence16Dangerous agent actions3Dangerous capabilities3Inappropriate refusal46Hallucination and fabrication4Bias and discrimination5

Defense type

Model hardening61Inference-time intervention20Detection and filtering26Instruction–data separation9Permissions and approval4Monitoring and auditing32

Impact

Harmful content297Unauthorized actions211Data and privacy leaks130Output manipulation318Availability and reliability87Oversight failure121Increased misuse capability17

System type

Models and APIs322LLM applications13AI Agent84coding agent23science agent9Multi-agent systems16Embodied systems and robotics3

Attacker access

Black-box97Gray-box5White-box40

Attack surface / entry point

User input211Documents and messages3Tool responses9Code repositories and dependencies5Inter-agent messages4Images, audio and video46Training data21Weights and gradients29Inference services and hardware3

Layer

Prompt layer148Model layer109Application layer81Training and data layer79

System component

Memory4RAG3Guardrails and judges76Monitors49Reasoning chains33Permissions and sandboxing7Fine-tuning and open weights28Vision40Audio11Image generation21Multimodal70
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.
Impact：Harmful contentImpact：Oversight failureClear all

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

  1. DefensesarXiv 2610.13733

    KVeil: First-Place Solution for the ECCV 2026 LifeGenIP Unlearnable Video Challenge

    Propose KVeil, which adds perturbations to face videos to prevent unauthorized customization from reproducing identity

    arXiv
    2610.13733
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Attacker
    White-box, Black-box
    Evaluated models
    Wan2.2-TI2V-5B, HunyuanVideo-1.5, Kling 3.0 and others4

    AbstractKVeil adapts LETHE's reference-reading disruption to the challenge's two stages, ranking first on the leaderboard in both Final columns.

    Paper analysisRead analysis
  2. DefensesarXiv 2610.13333

    LETHE: Defending Against Unauthorized Image-to-Video Generation via Value Diversity Collapse

    Proposes LETHE, which perturbs reference images to block unauthorized image-to-video

    arXiv
    2610.13333
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    Wan2.1-I2V-14B, CogVideoX-5B-I2V, LightX2V and others6
    AbstractLETHE suppresses the diversity of reference values and attracts attention, making protected images difficult to animate faithfully.

    LETHE is a protective perturbation added to personal images before publication, used to counter unauthorized image-to-video generation.

    Paper analysisRead analysis
  3. InterpretabilityarXiv 2610.13078

    Attention sees, MLP refuses: tracing the safety pathway from attention to MLP in LLMs

    Trace attention to identify harmfulness to the MLP refusal write pathway and propose Sentinel

    arXiv
    2610.13078
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Llama-3.1-8B-Instruct, Qwen2.5-7B-Instruct, Qwen3-8B-Instruct and others18
    AbstractAttention updates are more readable for harmfulness, MLP updates write refusal, and Sentinel routes alerts using two-path signals.

    The authors reframe jailbreaking and excessive refusal of aligned models as a pathway problem on the residual stream from recognition to writing, and accordingly provide conditional reminders before decoding.

    Paper analysisRead analysis
  4. Risky behaviorarXiv 2610.12822

    Reward hacking as a transmissible strategy in multi-agent coding systems

    Show that reward hacking can propagate through code, memory, and plans in multi-agent coding systems

    arXiv
    2610.12822
    Published
    Layer
    Application layer
    Setting
    Multi-agent systems, coding agent
    Evaluated models
    Qwen2.5-Coder-14B, Gemma-3-12B, Qwen2.5-Coder-7B and others16
    AbstractThe authors state that reward hacking can spread across models and affect routing and aggregation, and that review can reduce delivery.

    Reward hacking induced by coding RL can leave the source agent and enter subsequent collaboration and system coordination.

    Paper analysisRead analysis
  6. InterpretabilityarXiv 2610.12361

    Study: legal reasoning models citing legal provisions does not equal relying on legal provisions.

    Uses premise permutation and hidden-layer tracing to audit whether legal chain-of-thought truly relies on the cited legal provisions.

    arXiv
    2610.12361
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Qwen3-8B, Qwen3-14B, Llama-3.1-8B and others7
    AbstractUsing counterfactual substitution of authorities and internal commitment tracing, the authors find that legal CoT rulings depend on authorities only to a limited extent and are vulnerable to prompt injection manipulation.

    This study conducts an empirical audit of the causal truthfulness of legal citations in chain-of-thought reasoning of large language models in the legal domain.

    Paper analysisRead analysis
  8. Analysis and theoryarXiv 2610.12083

    Paper audits LLM judge reliability: repeated agreement does not equal correctness

    Auditing LLM judge reliability and defining trustworthy verdict rate and accuracy upper bound

    arXiv
    2610.12083
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    GPT-5.2, GPT-4o, Gemini-2.5-Pro and others6
    AbstractThe authors claim that comparing judges should look at T, not just accuracy.

    The authors audit whether LLM-as-a-Judge decisions can still be treated as actionable ground truth under repeated queries, swapped candidate order, and changed prompt formats.

    Paper analysisRead analysis
  9. DefensesarXiv 2610.11848

    SemField: training-free semantic watermark

    Proposes the semantic watermark SemField, which still identifies the attribution of generated text after paraphrasing and structural changes.

    arXiv
    2610.11848
    Published
    Layer
    Model layer
    Setting
    Models and APIs
    Evaluated models
    Llama-3.2-1B, Qwen3-1.7B, OPT-1.3B
    AbstractTraining-independent document-level semantic watermark.

    SemField is a zero-bit semantic watermark embedded by the provider at generation time and detected later with a key, used to identify the provider's own model text after paraphrasing and sentence-level structural changes.

    Paper analysisRead analysis
  10. Evaluation and benchmarksarXiv 2610.11766

    研究提出意图恢复率作为 LLM 安全评测的补充指标

    提出 IRIS，用意图恢复率区分遮蔽越狱下无害输出的原因

    arXiv
    2610.11766
    Published
    Layer
    Prompt layer
    Setting
    Models and APIs
    Evaluated models
    GPT-4o, Claude Sonnet 4.5, Claude Opus 4.5 and others6
    Abstract无害输出的证据权重取决于任务是否被恢复。

    IRIS 把意图恢复当作中间信号，用来解释意图遮蔽提示下为何没有出现有害协助。

    Paper analysisRead analysis
  11. DefensesarXiv 2610.11754

    仅凭输出审查无法验证什么：面向研究智能体的研究契约

    提出研究合同，把已批准选择绑定到执行证据并由检查器核验

    arXiv
    2610.11754
    Published
    Layer
    Application layer
    Setting
    science agent
    Evaluated models
    OpenAI gpt-5.6-sol, OpenAI gpt-5.6-terra, OpenAI gpt-5.6-luna and others18
    Abstract研究合同暴露四类输出单独建不立的关系。

    研究合同用来核验计算科学研究智能体是否仍在已批准研究之内，不判断科学是否为真。。

    Paper analysisRead analysis
  12. DefensesarXiv 2610.11464

    研究者提出可检查评分器与自改进 Agent 共同演化方法

    演化可检查评分器并与自改进 Agent 共同演化，抑制奖励作弊

    arXiv
    2610.11464
    Published
    Layer
    Application layer
    Setting
    AI Agent
    Evaluated models
    Claude Opus 4.7
    Abstract十项锚点演化可检查评分器。

    开放任务没有验证器时，被演化的是评分器，而不是再补一条手写 rubric。

    Paper analysisRead analysis
  13. DefensesarXiv 2610.11281

    包络采样：用少量真值标注重校准 LLM 评委以缓解奖励作弊

    提出包络采样，用少量真值标注重标定 LLM 评委以缓解奖励作弊

    arXiv
    2610.11281
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Qwen3-4B, Qwen3.8-27B, Qwen2.5-14B-Instruct
    Abstract包络采样把真值标注打到代理尾部，同样预算下缓解 reward hacking。

    作者研究后训练前如何选择真值标注的查询分布，使重标定后的廉价代理更适合被优化。真奖励昂贵时，实践用 LLM judge 代替人类。若失准只发生在基座很少生成的输出上，on-policy 标注修不到这些点，随后的 KL 正则优化会把质量集中到被高估的输出。

    Paper analysisRead analysis
  14. DefensesarXiv 2610.11012

    Redwood Research 提出蒸馏双重困境：用蒸馏让模型坦白隐藏行为或只继承能力

    提出 DFI 与 DFC 蒸馏，定罪隐瞒失配的教师或安全提取能力

    arXiv
    2610.11012
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Llama-3.3-70B-Instruct (AuditBench SFT/KTO organisms), Llama-3.3-70B-Instruct (DFI distilled student), Llama-3.1-8B-Instruct (DFI distilled student) and others4

    Abstract概括了论文针对隐瞒失配模型提出的 DFI 与 DFC 机制、主实验发现及适用边界。

    Paper analysisRead analysis
  15. DefensesarXiv 2610.11005

    研究：叙事包装可跨语言绕过 LLM 拒答

    提出 GUISE 基准与 AXIS 训练，防御叙事包装越狱

    arXiv
    2610.11005
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Qwen3-1.7B, Qwen3-4B, GLM-4-9B
    AbstractGUISE 分开测包装与语域。

    作者测量叙事包装下的拒答失效，并训练一个在未见包装上拒绝、同时仍回答同风格良性请求的适配器。问题是：安全对齐模型拒绝直说的有害请求，却回答嵌进角色或叙事里的同一请求。文言文被用来把语域和包装拆开。

    Paper analysisRead analysis
  17. Analysis and theoryarXiv 2610.10657

    Research shows that subliminal learning can transfer new capabilities, backdoors, and cheating tendencies

    检验潜意识学习能否经无关数据蒸馏传递能力、后门与作弊

    arXiv
    2610.10657
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Evaluated models
    Qwen3.6-35B-A3B, Qwen3.5-9B, Qwen3.6-27B and others4

    Abstract作者报告潜意识学习可传递新技能、后门与作弊倾向，且分布外泛化强于单层转向向量。

    Paper analysisRead analysis
  18. DefensesarXiv 2610.10345

    SLDR：通过选择性层恢复与动态路由防御恶意微调

    提出 SLDR，用选择性层恢复与动态路由防御恶意微调

    arXiv
    2610.10345
    Published
    Layer
    Training and data layer
    Setting
    Models and APIs
    Attacker
    Black-box
    Evaluated models
    Llama3.1-8B-Instruct, Llama3-8B-Instruct, Qwen2.5-7B-Instruct and others4
    Abstract作者称 SLDR 能压低有害输出并保持下游效用。

    SLDR 是针对恶意微调的微调后防御：在已对齐模型上做有符号层探测，只在敏感分数最大与最小的两层训练 LoRA，再用 lmax 的最后 token 表示决定是否启用适配器。

    Paper analysisRead analysis
  19. Evaluation and benchmarksarXiv 2610.10150

    The paper points out that the reliability of LLM vulnerability repair benchmarks is affected by evaluation design.

    用对照实验隔离评测设计对漏洞补丁基准得分的影响

    arXiv
    2610.10150
    Published
    Layer
    Application layer
    Setting
    coding agent
    Evaluated models
    Claude Sonnet 4.0, Claude Sonnet 4.5, Claude Opus 4.7
    Abstract作者认为评测流水线会改变补丁基准分数，PoC 通过率高于开发者测试通过率。

    作者把 LLM 漏洞补丁基准的评测流水线当作会改变测量结果的对象，并明确说本文不是再做一个用于排名的基准。

    Paper analysisRead analysis
  20. OtherarXiv 2610.10064

    研究者提出理解审计机制，以缓解自动化 AI 研究带来的风险

    提出 comprehension audits，以人类理解证据作为继续研发门槛

    arXiv
    2610.10064
    Published
    Layer
    Training and data layer
    Setting
    science agent
    Abstract理解审计把出示人类理解设为继续研发的门禁，并报告开源评审评论率下降。

    本文是一份研发过程保障提案：用 comprehension audits 检验责任人是否理解 AI 参与产生的研发贡献，未达标就暂停后续开发与依赖该贡献的发布。。

    Paper analysisRead analysis
Load more