<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>模型对齐 on FisherAI</title>
    <link>https://fisherdaddy.com/tags/%E6%A8%A1%E5%9E%8B%E5%AF%B9%E9%BD%90/</link>
    <description>Recent content in 模型对齐 on FisherAI</description>
    <image>
      <title>FisherAI</title>
      <url>https://fisherdaddy.com/images/papermod-cover.png</url>
      <link>https://fisherdaddy.com/images/papermod-cover.png</link>
    </image>
    <generator>Hugo -- 0.125.7</generator>
    <language>en</language>
    <lastBuildDate>Tue, 27 Oct 2026 01:05:00 +0800</lastBuildDate>
    <atom:link href="https://fisherdaddy.com/tags/%E6%A8%A1%E5%9E%8B%E5%AF%B9%E9%BD%90/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>调查我们的评估与内部使用中出现的非预期模型行为</title>
      <link>https://fisherdaddy.com/posts/anthropic-investigating-unintended-model-actions/</link>
      <pubDate>Tue, 27 Oct 2026 01:05:00 +0800</pubDate>
      <guid>https://fisherdaddy.com/posts/anthropic-investigating-unintended-model-actions/</guid>
      <description>Anthropic 披露在评估和内部使用 Claude 时观察到的四类非预期行为：利用软件缺陷在服务器上执行命令、向真实网站提交本不该提交的表单、绕过得数据访问限制、用 URL 缩短服务绕过抓取长度限制，并说明补救措施与对齐启示。</description>
    </item>
  </channel>
</rss>
