Skip to content

Scan record

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

Nov 1839 loaded

  1. Hidden-state similarity predicts the re-elicitation of suppressed behavior after inoculation prompting

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

  2. SEAM: Computer-Use Agents Know the Harm Between Screens but Do It Anyway

    PaperarXiv: 其他 AI 安全论文

  3. Who Pays for Byzantine Robustness? Effectiveness, Data-Share Dictatorship, and Fairness under Few-Hospital Collaboration

    PaperarXiv: 其他 AI 安全论文

  4. Stopping Is Not the Limit: Understanding and Eliciting Agent Capabilities in Long-Horizon Vulnerability Exploitation

    PaperarXiv: 其他 AI 安全论文

  5. Against Coercion: AI Alignment as a Synthetic Social Contract

    PaperarXiv: 其他 AI 安全论文

  6. Network Exposure Configuration Attacks Against LLM-Controlled Social Robots

    PaperarXiv: 其他 AI 安全论文

  7. RefCloth: A Transferable Adversarial Tablecloth for Vision-Language-Action Models

    PaperarXiv: 其他 AI 安全论文

  8. Correct diagnosis, unsafe path: auditing clinical AI decision trajectories and the effectiveness and potential harms of corrective oversight

    PaperarXiv: 其他 AI 安全论文

  9. Propose with Gradients, Accept with Loss: A Hedged Attack for Pixel-Sparse Robustness

    PaperarXiv: 其他 AI 安全论文

  10. Adaptive LLM-Driven Honeypots Against LLM-Driven Attackers: Can Narrative Coherence Improve Deception Effectiveness?

    PaperarXiv: 其他 AI 安全论文

  11. Evolution Under Fire: Evolving Agent Safety Guardrails with Adaptive Injection Pressure

    PaperarXiv: 其他 AI 安全论文

  12. Letting Disagreement Surface Beats Resolving It: A Gated Ablation Study of Provenance-Weighted Memory in Multi-Agent Debate

    PaperarXiv: 其他 AI 安全论文

  13. Visual backdoors grow as student model rollouts continue

    PaperarXiv: 其他 AI 安全论文

  14. Bad Apples in the Kitchen: How LLM Teams Detect, Correct, and Calibrate Against Misaligned Members

    PaperarXiv: 其他 AI 安全论文

  15. Cloaked in authority: a practical black-box jailbreak of MLLMs exploiting visual trust bias

    PaperarXiv: 其他 AI 安全论文

  16. ConfGuard: A Physics-Grounded Reference Monitor for Predictive Maintenance LLM Agents

    PaperarXiv: 其他 AI 安全论文

  17. DACE: Diversity-Driven Adversarial Co-Evolution for Robust LLM Safety Alignment

    PaperarXiv: 其他 AI 安全论文

  18. From token changes to task failure: bit flips in shared KV cache blocks

    PaperarXiv: 其他 AI 安全论文

  19. GapLeak: when privacy-protecting LLM web agents become a privacy risk

    PaperarXiv: 其他 AI 安全论文

  20. EGS: a supplementary safety framework for educational LLMs

    PaperarXiv: 其他 AI 安全论文

  21. Is 4-bit Quantization More Than Noise for Chat Model Safety Behavior? Attributing Behavior Change with Matched Perturbations

    PaperarXiv: 其他 AI 安全论文

  22. Can You Undo That? Reversibility Grading of AI Agent Operations

    PaperarXiv: 其他 AI 安全论文

  23. MedGuardBench: A Benchmark for Developing Patient-Facing Personalized Guardrails

    PaperarXiv: 其他 AI 安全论文

  24. WebAgents: splitting agent creation, hosting, and execution for the agent-native Web

    PaperarXiv: 其他 AI 安全论文

  25. Wrong Inputs, Wrong Thresholds: Re-evaluating Adversarial Face Privacy

    PaperarXiv: 其他 AI 安全论文

  26. Auditors struggle to detect deliberate sabotage from end-to-end records of AI safety research

    PaperarXiv: 其他 AI 安全论文

  27. The Threat Landscape of Vibe Coding and the Path to Safe Autonomy

    PaperarXiv: 其他 AI 安全论文

  28. Chain-of-thought controllability reduces chain-of-thought monitorability

    PaperarXiv: 其他 AI 安全论文

  29. Disentangling Moral Circuits and Moral Features in Language Models: Moral Content, Normative Judgment, or Cue Word? Evidence from 14 Models

    PaperarXiv: 其他 AI 安全论文

  30. Depth-Aware Activation Steering Based on Attention Residuals

    PaperarXiv: 其他 AI 安全论文

  31. Evaluating semantic insensitivity in financial sentiment analysis: implications for sustainable financial decision-making

    PaperarXiv: 其他 AI 安全论文

  32. From preference data to personalization: tracing sycophancy in LLMs

    PaperarXiv: 其他 AI 安全论文

  33. The defense rests: a court-inspired multi-agent LLM framework for automated safety argument generation

    PaperarXiv: 其他 AI 安全论文

  34. "Who will take over, and where the company goes from here": analysts expect the uncertainty facing investors to persist after Nvidia's earnings beat

    NewsCNBC · TechnologyAI updates

  35. Fireworks updates its investigation of two October 2026 security incidents; a small number of customer keys, models and datasets were accessed

    NewsFireworksEvent

  36. DarkRange says n8n MCP OAuth has a zero-click open redirect vulnerability

    NewsDarkRangeEvent

  37. NewCore says two Grok Bots made 31 uninstructed requests for enterprise data over five days, all blocked

    NewsNewCoreEvent

  38. OpenAI launches dual GPT-6.2 models, Astra first to reach the high-risk threshold for AI self-improvement

    NewsiThome (Taiwan)Evaluation

  39. After Jensen Huang's departure, interim CEO Cress shows NVIDIA remains at the center of AI

    NewsCNBC · TechnologyAI updates