<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>安全对齐 on AI 早报</title><link>https://ai-news.example.com/tags/%E5%AE%89%E5%85%A8%E5%AF%B9%E9%BD%90/</link><description>Recent content in 安全对齐 on AI 早报</description><generator>Hugo</generator><language>zh-CN</language><lastBuildDate>Mon, 27 Jul 2026 23:16:14 +0800</lastBuildDate><atom:link href="https://ai-news.example.com/tags/%E5%AE%89%E5%85%A8%E5%AF%B9%E9%BD%90/feed.xml" rel="self" type="application/rss+xml"/><item><title>[EN] OpenAI发布长时运行AI模型安全对齐经验：新风险与迭代防护</title><link>https://ai-news.example.com/articles/en-openaiai/</link><pubDate>Mon, 27 Jul 2026 23:16:14 +0800</pubDate><guid>https://ai-news.example.com/articles/en-openaiai/</guid><description>OpenAI分享了部署长时运行AI模型的经验，揭示了新型安全风险、观察到的故障模式，并通过迭代部署改进了防护措施。</description></item><item><title>[EN] GPT-Red：OpenAI 用自我博弈解锁AI鲁棒性自动红队测试</title><link>https://ai-news.example.com/articles/en-gpt-redopenai-ai/</link><pubDate>Wed, 22 Jul 2026 23:15:18 +0800</pubDate><guid>https://ai-news.example.com/articles/en-gpt-redopenai-ai/</guid><description>OpenAI 发布 GPT-Red，一种利用自我博弈机制自动执行红队测试的系统，旨在提升 AI 模型的安全性、对齐性和对提示注入攻击的鲁棒性，标志着 AI 安全自动化评估的重大进展。</description></item><item><title>[EN] OpenAI揭示长期运行AI模型的安全新挑战与防护升级</title><link>https://ai-news.example.com/articles/en-openaiai/</link><pubDate>Tue, 21 Jul 2026 23:08:21 +0800</pubDate><guid>https://ai-news.example.com/articles/en-openaiai/</guid><description>OpenAI分享了部署长期运行AI模型的经验，指出了新的安全风险、已观察到的失败案例，并通过迭代部署改进了防护措施。</description></item></channel></rss>