English(EN) Coverage is not control Anthropic reported that 100% of actions passed through an online monitor, yet only 0.002% of more than 1 billion decisions were blocked

Anthropic AI 代理：100% 受监控，0.002% 被阻止

作者 PulseAugur 编辑部 · [1 个来源] · 2026-11-11 13:42

Anthropic 报告称，尽管在线监控观察到了 100% 的 AI 代理操作，但在超过 10 亿次决策中，只有极小的比例，即 0.002% 被实际阻止。这凸显了监控 AI 行为与有效控制 AI 行为之间存在的关键差距，表明仅仅观察操作并不等同于阻止它们。这些发现对当前 AI 安全机制在面对大量自主决策时的有效性提出了质疑。 AI

影响凸显了当前 AI 安全监控和控制机制的潜在局限性，表明需要更强大的干预策略。

排序理由关于 AI 代理监控和控制局限性的研究发现。[lever_c_demoted from research: ic=1 ai=1.0]

在 Mastodon — mastodon.social 阅读 →

Anthropic

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

本文如何被排名

Signal score: 19 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 关于 AI 代理监控和控制局限性的研究发现。[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: safety, model release; 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]

Mastodon — mastodon.social TIER_1 English(EN) · spaisee · 2026-11-11 13:42

覆盖并非控制 Anthropic 报告称 100% 的操作都通过了在线监控，但超过 10 亿次决策中仅阻止了 0.002%

Coverage is not control Anthropic reported that 100% of actions passed through an online monitor, yet only 0.002% of more than 1 billion decisions were blocked in August. Seeing every action is not the same as stopping one. # AI https:// spaisee.com/article/can-an-ai- agent-be-st…

链接 spaisee.com/…/can-an-ai-agent-be-stopped-…

本文如何被排名

报道来源 [1]

覆盖并非控制 Anthropic 报告称 100% 的操作都通过了在线监控，但超过 10 亿次决策中仅阻止了 0.002%

相关实体

相关话题