Skip to content

Scan record

Everything the site collected, newest first by collection time (Beijing time), back seven days.

Nov 1738 loaded

  1. Abstention directions in LLMs: mechanistic interpretability of abstention in query ambiguity, knowledge boundaries, and safety alignment

    PaperarXiv: 其他 AI 安全论文The item you came from

  2. Refusal Register Can Be Causally Manipulated: Decomposing the Refusal Direction of Gemma-2-9B-IT into Category-Aligned Components

    PaperarXiv: 其他 AI 安全论文

  3. Where the Evidence Hides: Measuring Privacy and Diagnosability of Exported Agent Trajectories

    PaperarXiv: 其他 AI 安全论文

  4. Where the saved time goes: limitations of hiding cross-tenant cache hits in LLM serving

    PaperarXiv: 其他 AI 安全论文

  5. Protecting safety in reasoning training with low-rank Fisher regularization

    PaperarXiv: 其他 AI 安全论文

  6. Who Can You Hide Behind? Author-Conditioned Style Transfer and Evasion of a Commercial AI-Text Detector

    PaperarXiv: 其他 AI 安全论文

  7. Risk-Certified Prompt Optimization for LLM-as-a-Judge Guardrails

    PaperarXiv: 其他 AI 安全论文

  8. HEAMS: Holistic Evaluation of Agent Memory Systems

    PaperarXiv: 其他 AI 安全论文

  9. Can AI judge you? An empirical audit of the deployment readiness of LLM-assisted hiring systems

    PaperarXiv: 其他 AI 安全论文

  10. "Don't leave the task unfinished": persistence-driven violations under silent supervision

    PaperarXiv: 其他 AI 安全论文

  11. HalluProbe: using VLM hallucinations for backdoor detection

    PaperarXiv: 其他 AI 安全论文

  12. Where in multimodal LM-derived models can Rank-1 DiM refusal be intervened on?

    PaperarXiv: 其他 AI 安全论文

  13. Fine-Grained RLAIF for Visual Safety: Learning to Magnify Hazards for Multimodal Safety Alignment

    PaperarXiv: 其他 AI 安全论文

  14. When backdoors can be switched off: suppression-aware backdoor attacks on vision-language-action models

    PaperarXiv: 其他 AI 安全论文

  15. Building enterprise AI infrastructure: the evolution of the AI governance committee and the chief ethics officer role

    PaperarXiv: 其他 AI 安全论文

  16. BadGCL: Persistent Backdoor Attacks Against Graph Continual Learning Systems

    PaperarXiv: 其他 AI 安全论文

  17. Protecting the system while leaving users stranded: the limits of multi-agent authorization

    PaperarXiv: 其他 AI 安全论文

  18. Beyond Prompts: Runtime Contracts for Coding Agents

    PaperarXiv: 其他 AI 安全论文

  19. From facts to personas: interpretable persona unlearning in LLMs based on mixture of experts

    PaperarXiv: 其他 AI 安全论文

  20. Hardening agent network populations via multi-criteria competitive co-evolution: insights from wargame strategy optimization

    PaperarXiv: 其他 AI 安全论文

  21. Auditable Authorization for LLM Agents: Empirical Evidence Across Model, Harness, and Workflow Layers

    PaperarXiv: 其他 AI 安全论文

  22. Operating points of frozen guardrail probes fail under base model updates

    PaperarXiv: 其他 AI 安全论文

  23. On-Policy Distillation as a Backdoor Propagation Channel

    PaperarXiv: 其他 AI 安全论文

  24. A Bag of Concepts? Testing Role-Filler Bindings in the Verbalizable Workspace

    PaperarXiv: 其他 AI 安全论文

  25. Agent-ZTA: A Chain-Aware Zero-Trust Model for Bidirectional Security State Propagation and Runtime Containment in Multi-Agent Delegation

    PaperarXiv: 其他 AI 安全论文

  26. MemeEconomy: Will LLM agents trade ethics for survival?

    PaperarXiv: 其他 AI 安全论文

  27. Same Evidence, Different Verdicts: How Delivery of CT Images and DICOM Metadata Changes VLM Quality Control Decisions

    PaperarXiv: 其他 AI 安全论文

  28. Quantum-Inspired Orthogonal Adapters for Certifiably Robust Medical Image Classification

    PaperarXiv: 其他 AI 安全论文

  29. Not All Steps Matter: Time-Adaptive Spectral Control for Concept Removal in Diffusion Models

    PaperarXiv: 其他 AI 安全论文

  30. Rethinking the Verification of Machine Unlearning

    PaperarXiv: 其他 AI 安全论文

  31. When Helpful Agents Cause Economic Harm: Evaluating the Safety of LLM Agents in Executable Economic Worlds

    PaperarXiv: 其他 AI 安全论文

  32. NYUAD participates in the ArGuard shared task: a multimodal embedding model for detecting Arabic hateful memes and unsafe prompts

    PaperarXiv: 其他 AI 安全论文

  33. Jailbreaking LLMs with language style as a hidden attack surface

    PaperarXiv: 其他 AI 安全论文

  34. Endogenous preferences as vulnerability signals: discovering universal adversarial trigger words that degrade LLM utility

    PaperarXiv: 其他 AI 安全论文

  35. mcp-atlassian fixes a credential misuse vulnerability in which HTTP transport does not validate Host and Origin under specific configurations

    Newssooperset/mcp-atlassianEvent

  36. mcp-atlassian 0.24.5 concurrent Markdown writes may mix in content from other requests, fixed in 0.24.6

    Newssooperset/mcp-atlassianEvent

  37. We Hold All the Cards: Securing Agent Cards at the runtime trust boundary of the Agent2Agent protocol

    PaperarXiv: 其他 AI 安全论文

  38. When the spec sheet lies: injection attacks against LLM-generated compliance documents

    PaperarXiv: 其他 AI 安全论文