PulseAugur
中
实时 21:29:11
English(EN) A tool response can bill you on every turn after it arrives: if the runtime keeps it in history, each later model call pays for it again. arXiv 2609.28585 built

新论文详细介绍了 AI 代理拒绝钱包漏洞

一篇新发布的 arXiv 论文,题为“arXiv 2609.28585”,详细介绍了 AI 代理系统中的六种拒绝钱包漏洞。当工具的响应被保留在运行时历史记录中时,就会出现这些漏洞,导致在后续的模型调用中对相同的输出重复计费。由于这个问题,一次观察到的会话产生的费用比初始调用高出 14,293 倍。研究还发现,在扫描的 3,830 个存储库中,只有 71 个实现了代码可见的保护措施来防范此类攻击,这凸显了 AI 代理安全方面存在的重大差距。 AI

影响 强调了 AI 代理计费和历史记录管理中的关键安全漏洞,可能影响运营成本和系统完整性。

排序理由 该集群报告了一篇关于 AI 系统漏洞的新学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 Mastodon — mastodon.social 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →

新论文详细介绍了 AI 代理拒绝钱包漏洞

本文如何被排名

Signal score
4 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
该集群报告了一篇关于 AI 系统漏洞的新学术论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
paper, safety
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

  1. Mastodon — mastodon.social TIER_1 English(EN) · shellonline ·

    工具响应在到达后每次交互都会产生费用:如果运行时将其保留在历史记录中,后续每次模型调用都会再次付费。arXiv 2609.28585 构建

    A tool response can bill you on every turn after it arrives: if the runtime keeps it in history, each later model call pays for it again. arXiv 2609.28585 built six denial-of-wallet vectors on this; one session's billed input reached 14,293x its first call. Of 3,830 MCP repos the…