<?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>AI Safety on AmeBob</title>
    <link>https://blog.amebob.cn/categories/ai-safety/</link>
    <description>Recent content in AI Safety on AmeBob</description>
    <image>
      <title>AmeBob</title>
      <url>https://blog.amebob.cn/%3Clink%20or%20path%20of%20image%20for%20opengraph,%20twitter-cards%3E</url>
      <link>https://blog.amebob.cn/%3Clink%20or%20path%20of%20image%20for%20opengraph,%20twitter-cards%3E</link>
    </image>
    <generator>Hugo -- 0.148.1</generator>
    <language>en-us</language>
    <lastBuildDate>Sun, 06 Sep 2026 15:29:44 +0800</lastBuildDate>
    <atom:link href="https://blog.amebob.cn/categories/ai-safety/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>jailbreak-new-dir</title>
      <link>https://blog.amebob.cn/posts/post-20260906-152944/</link>
      <pubDate>Sun, 06 Sep 2026 15:29:44 +0800</pubDate>
      <guid>https://blog.amebob.cn/posts/post-20260906-152944/</guid>
      <description>本文讨论了通过伪造对话历史和API调用进行jailbreak攻击的方法，而无需搜索和包装。微软指出，这种“破绽”的原因是会话本地存储，而chatgpt的云端会话存储则无此问题，同时开源LLM必将长久面临此类问题。微软提出了针对盈利的LLM提供商的解决建议，包括对会话信息进行签名、保留云端会话等。</description>
    </item>
  </channel>
</rss>
