The BasePeople

Shilong Pan

PDL Lab, National University of Defense Technology

Titles as the sources gave them. In 1 entry, with 3 checked quotations.

What they said

Newest first. Each quotation was found word for word in the source.

“By constraining tail risk rather than the overall average, TAS suppresses rare high-cost failures while preserving utility in normal cases.”
Tail-risk-aware Agent Safety: Constraining Tail Risks Across Multi-step LLM Agent WorkflowsarXiv, November 7, 2026. Speaking as: Other. The paper's abstract, stating what the proposed TAS framework is meant to achieve.
“Since most agent behaviors are typically safe while a small fraction may incur severe failures, an overall-average risk measure can mask rare but severe risks.”
Tail-risk-aware Agent Safety: Constraining Tail Risks Across Multi-step LLM Agent WorkflowsarXiv, November 7, 2026. Speaking as: Other. The introduction, explaining why the authors measure worst-case rather than average risk.
“Removing the shared budget constraint (w/o constraint) causes the largest safety degradation, showing that the budget is critical for controlling risk propagation.”
Tail-risk-aware Agent Safety: Constraining Tail Risks Across Multi-step LLM Agent WorkflowsarXiv, November 7, 2026. Speaking as: Other. From the ablation study, where each component of TAS was removed in turn.

In the Base

Every entry they appear in.

  1. ResearcharXivWrote it

    Tail-risk-aware Agent Safety: Constraining Tail Risks Across Multi-step LLM Agent Workflows

    Researchers at the National University of Defense Technology propose TAS, a defence for LLM agents that treats a workflow as linked steps sharing one risk budget, aimed at rare but severe failures. They report lower attack success than eleven baselines on three benchmarks.