EvoSafeHarness: Evolving Model- and Domain-Specific Harnesses for Securing Agents

  • 类型：arxiv
  • 标识：2609.05903
  • 链接：https://arxiv.org/abs/2609.05903
  • 主分类：evaluation
  • 形态：application
  • 被引：0
  • 被引来源：OpenAlex
  • OpenAlex被引：0
  • TLDR：Large Language Model (LLM) agents are turning language into real-world effects, making safety necessary against both indirect prompt injections and direct harmful requests. System-level safety harnesses add an enforcement layer beyond model-level defenses, but existing harnesses are usually designed once by experts and applied across heterogeneous models and domains. Effective protection is deployment-dependent: models differ in how much enforcement they need before utility declines, while domains differ in the effects, state, and action sequences that must be governed. A harness that is stric
  • OpenAlex ID：W7212001064
  • OpenAlex DOI：10.48550/arxiv.2609.05903
  • DOI：10.48550/arxiv.2609.05903
  • DOI来源：OpenAlex
  • 开放获取：green
  • 开放获取链接：https://arxiv.org/pdf/2609.05903
  • OpenAlex更新：2026-09-12
  • 副分类：agent
  • 待LLM分类：否
  • 标题中文：EvoSafeHarness：面向 Agent 安全防护的演进式模型与领域专属 Harness
  • TLDR中文：大语言模型（LLM）agent 正将语言转化为现实世界影响，因此必须同时防御间接提示注入与直接有害请求，安全防护不可或缺。系统级安全 harness 在模型级防御之上增加了一层强制执行机制，但现有 harness 通常由专家一次性设计后便通用于异构模型与领域。有效的保护具有部署依赖性：不同模型在效用下降前所能承受的强制程度不同，不同领域在需要被管控的影响、状态与动作序列上也各不相同。过于严苛的 harness……
  • 来源文件：
  • /inbox/tom/_candidates/2026-09-11-agent-rag-longcontext-candidates.json
  • [OpenAlex backfill]