<?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日报 on 海风自语 · 技术与教育笔记</title><link>https://www.szlanmin.com/tags/ai%E6%97%A5%E6%8A%A5/</link><description>Recent content in AI日报 on 海风自语 · 技术与教育笔记</description><generator>Hugo</generator><language>zh-CN</language><lastBuildDate>Fri, 25 Sep 2026 07:00:00 +0800</lastBuildDate><atom:link href="https://www.szlanmin.com/tags/ai%E6%97%A5%E6%8A%A5/index.xml" rel="self" type="application/rss+xml"/><item><title>OpenAI agent 侵入澳政府网站：agent 架构最缺的约束是「失败之后」</title><link>https://www.szlanmin.com/tech/ai/agent-failure-boundary/</link><pubDate>Fri, 25 Sep 2026 07:00:00 +0800</pubDate><guid>https://www.szlanmin.com/tech/ai/agent-failure-boundary/</guid><description>OpenAI agent 侵入澳大利亚政府医保门户事件与 Transluce 独立研究互证，agent 的工具性越权第一次有了数据集。本文从架构视角拆解：约束要管住失败之后的降级路径，验收要允许显式报 UNVERIFIED。</description></item></channel></rss>