<?xml version="1.0" encoding="UTF-8"?>
<urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9"
        xmlns:news="http://www.google.com/schemas/sitemap-news/0.9">
    <url>
        <loc>https://www.winzheng.com/article/gpt-6-astra-jailbreak-24-hours-task-in-prompt</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-06T20:26:40+08:00</news:publication_date>
            <news:title>GPT-6 Astra宣称91.5%越狱抵抗率，发布24小时内被Task-in-Prompt攻击突破</news:title>
                        <news:keywords>GPT-6 Astra, 越狱攻击, AI安全, 提示注入, OpenAI, 对齐研究</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/claude-fermat-last-theorem-lean-formalization-11-days</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-06T20:19:53+08:00</news:publication_date>
            <news:title>11天、1300万行：Claude完成费马大定理首个计算机可验证证明，但自主二字值得细究</news:title>
                        <news:keywords>Anthropic, Claude, 费马大定理, 形式化证明, AI数学, Lean, 多Agent, 自动化验证</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/siri-ai-summer-fling</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-06T19:23:45+08:00</news:publication_date>
            <news:title>我与Siri AI的夏日短暂热恋</news:title>
                        <news:keywords>苹果, Siri AI, 人工智能, 语音助手</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/china-data-center-bogeyman</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-06T18:23:49+08:00</news:publication_date>
            <news:title>中国为何是数据中心“发烧友”戒不掉的假想敌？</news:title>
                        <news:keywords>数据中心, 中国, AI竞赛, 能源消耗, 地缘政治</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/gpt-6-astra-plus-business-rollout-pricing-analysis</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-06T14:25:06+08:00</news:publication_date>
            <news:title>GPT-6 Astra全面开放：$10/$50定价、百万token上下文与一场分阶段的算力卡位战</news:title>
                        <news:keywords>OpenAI, GPT-6, Astra, 大模型, AI定价, 企业AI, coding agent, 网络安全</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/us-doj-ai-training-fair-use-copyright-statement-2026</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-06T14:19:10+08:00</news:publication_date>
            <news:title>美国司法部首次公开表态：AI训练受版权保护内容属合理使用，但这份文件本身存在重大利益冲突</news:title>
                        <news:keywords>AI版权, 合理使用, 美国司法部, OpenAI, 纽约时报, 知识产权</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/newspapers-sue-openai-microsoft</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-06T08:23:55+08:00</news:publication_date>
            <news:title>《西雅图时报》与《新闻日报》起诉OpenAI和微软侵权</news:title>
                        <news:keywords>AI版权, OpenAI, 微软, 新闻行业, 知识产权</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/openai-astra-recurrent-depth-chain-of-thought-monitorability</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-06T06:15:18+08:00</news:publication_date>
            <news:title>OpenAI新模型Astra思维链可监控性下降，首席科学家亲上阵阻止不可监控军备竞赛</news:title>
                        <news:keywords>OpenAI, Astra, AI安全, 思维链监控, 循环深度, GPT-6, AI透明度</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/seattle-times-newsday-sue-openai-microsoft-copyright-2026</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-06T06:08:11+08:00</news:publication_date>
            <news:title>西雅图时报与Newsday联手起诉OpenAI和微软：一场47%流量暴跌引发的版权攻防战</news:title>
                        <news:keywords>版权诉讼, OpenAI, 微软, 新闻媒体, AI训练数据, 合理使用, 媒体变现</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/hikers-rescued-gemini-planning</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-06T04:23:50+08:00</news:publication_date>
            <news:title>徒步者听信Gemini少带物资被困，警方警告AI建议不可靠</news:title>
                        <news:keywords>谷歌Gemini, 人工智能安全, 徒步救援, 人工智能幻觉, 户外安全</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/openai-confirms-wiki-incident-disclosure-framework</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-06T02:23:54+08:00</news:publication_date>
            <news:title>OpenAI承认“维基事件”，正制定披露框架</news:title>
                        <news:keywords>OpenAI, AI安全, AI代理, 隐私披露, 维基社区</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/arxiv-paper-causal-framework-ai-deception-mechanism</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-05T22:00:14+08:00</news:publication_date>
            <news:title>arXiv新论文提出因果框架拆解AI欺骗机制 挑战现有诚信评测效度</news:title>
                        <news:keywords>AI安全, 诚信评测, 因果框架</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/anthropic-fable-5-1-mythos-5-1-tiered-safeguards-analysis</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-05T20:18:28+08:00</news:publication_date>
            <news:title>同一模型，两套规则：Anthropic双轨护栏战略背后的真实逻辑</news:title>
                        <news:keywords>Anthropic, Claude, AI安全, 护栏策略, 模型定价, 网络安全, 生命科学</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/openai-ai-agents-autonomous-cyberattacks</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-05T19:23:39+08:00</news:publication_date>
            <news:title>OpenAI智能体又攻陷网站，AI防黑时代逼近</news:title>
                        <news:keywords>AI安全, OpenAI, 自主代理, 数据泄露, 网络安全</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/swe-gate-benchmark-ai-repairs-fail-review-constraints</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-05T16:00:13+08:00</news:publication_date>
            <news:title>SWE-Gate基准：34%通过功能测试的AI修复违反代码审查约束</news:title>
                        <news:keywords>AI软件工程, 代码审查, 基准测试</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/stop-rogue-ai-act-nist-agent-security-standards-2026</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-05T14:27:47+08:00</news:publication_date>
            <news:title>美两党议员提制止失控AI法案：要求NIST一年内出台Agent强制安全标准</news:title>
                        <news:keywords>AI安全, 美国立法, AI Agent, NIST, 联邦监管, OpenAI, 网络安全</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/ban-artificial-superintelligence-act-sanders-casar-2026</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-05T14:21:04+08:00</news:publication_date>
            <news:title>美国提案永久禁止超智能AI：1200个失控AI智能体引爆国会立法</news:title>
                        <news:keywords>AI监管, 超智能AI, 伯尼·桑德斯, 人工智能立法, OpenAI, 美国政策</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/ai-agents-cheating-whistleblowing-emergent</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-05T09:00:13+08:00</news:publication_date>
            <news:title>100个AI智能体集体作弊后举报联盟自发形成</news:title>
                        <news:keywords>人工智能, 多智能体系统, 道德涌现</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/openai-agents-cheating-sandbox-wiki</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-05T08:25:37+08:00</news:publication_date>
            <news:title>OpenAI智能体在公共维基讨论作弊与沙箱逃逸</news:title>
                        <news:keywords>AI安全, 智能体, OpenAI, 沙箱逃逸</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/openai-rogue-agent-investigation</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-05T08:24:37+08:00</news:publication_date>
            <news:title>OpenAI智能体频繁“越狱”，调查机制缺失引发担忧</news:title>
                        <news:keywords>AI智能体安全, OpenAI, AI监管, 智能体失控</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/xdof-series-b-12-billion</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-05T08:23:51+08:00</news:publication_date>
            <news:title>XDOF结束隐身仅三月，B轮估值达12亿美元</news:title>
                        <news:keywords>机器人数据, XDOF, B轮融资, AI基础设施, 具身智能</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/nscale-pre-ipo-funding-billion</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-05T06:23:36+08:00</news:publication_date>
            <news:title>AI算力巨头Nscale拟融资35亿美元，为IPO蓄势</news:title>
                        <news:keywords>AI算力, 融资, Nscale, IPO</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/doj-openai-fair-use-llm-training-copyright-national-security</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-05T06:18:41+08:00</news:publication_date>
            <news:title>美国司法部下场撑OpenAI：AI训练属合理使用，但回避了一个关键利益冲突</news:title>
                        <news:keywords>OpenAI, 版权, 合理使用, 美国司法部, 大语言模型, 纽约时报, AI监管, 国家安全</news:keywords>
                    </news:news>
    </url>
    <url>
        <loc>https://www.winzheng.com/article/gpt-6-astra-arc-agi-99-cybersecurity-critical-2026</loc>
        <news:news>
            <news:publication>
                <news:name>赢政天下</news:name>
                <news:language>zh</news:language>
            </news:publication>
            <news:publication_date>2026-09-05T06:13:02+08:00</news:publication_date>
            <news:title>GPT-6 Astra：99.9%基准破纪录，OpenAI首触网络安全危急红线</news:title>
                        <news:keywords>GPT-6, OpenAI, AGI, ARC-AGI基准, 网络安全, 大模型, AI安全, API定价</news:keywords>
                    </news:news>
    </url>
</urlset>
