01快讯
OpenAI 就智能体越权访问澳大利亚政府网站致歉
The company also detailed how some of those breaches had happened, and outlined additional measures it is taking to assess the impact of the events.
来源链接可在原站核对查看原文 ↗2026-09-29
AI 快讯日报.
2026-09-29 收录的 20 条 AI 快讯，按采集时的来源记录回看。消息仍需核对原始发布页。
这期内容仍需回到原始发布页核对。
20 条精选查看其他日期 ↗
02快讯
AI 智能体安全公司 Reco 融资 5500 万美元，赛道竞争者密集
The round builds on a $30 million fundraise in February, taking the company's total funding to $140 million.
来源链接可在原站核对查看原文 ↗03快讯
DeepSeek Harness 桌面端预览版 v0.2 发布，支持插件安装与自动化任务
新版提供 macOS 和 Windows 安装包，新增插件管理页面与自动化任务功能，优化文件预览和代码变更展示。60% 用户已用第三方插件，官方将建插件市场。
来源链接可在原站核对查看原文 ↗04快讯
美国众议员 Ro Khanna 致信 DeepSeek、阿里、Moonshot AI，追问超级智能与 RSI 安全措施
Rep. Ro Khanna (D-CA) is asking Chinese AI companies and US intelligence officials to assess their own abilities to audit and respond to AI safety risks.
来源链接可在原站核对查看原文 ↗05快讯
Anthropic IPO 招股书自曝巨额亏损并警告 AI 灾难性风险
As Anthropic gears up for its public debut, a preview of the company’s own IPO filing reportedly doesn’t portray it in a positive light.
来源链接可在原站核对查看原文 ↗06快讯
实测 Claude Sonnet 5.5：代码动画断档领先 gpt6，但 3D 不如 Opus 5.5
要是OpenAI真的把200刀套餐额度砍半的话我只能忍痛用两个claude账号了
来源链接可在原站核对查看原文 ↗07快讯
荣耀 Magic9 全系获首批“AI 智能体手机”入网认证
荣耀 CEO 李健宣布 Magic9 全系机型获首批“AI 智能体手机”入网认证，标志 AI 智能体手机走向标准化。该系列昨日发布，共三款机型，4499 元起售，Pro Max 版配 2 亿主摄与 8800mAh 电池。
来源链接可在原站核对查看原文 ↗08快讯
从 Bag-of-Words 到 Jev：Sebastian Raschka 梳理语言模型文本分类方法史并实测 Jev
A Visual Guide to RNNs, CNNs, Transformers, and Calibration, with Hands-On Experiments on Accuracy and Efficiency
来源链接可在原站核对查看原文 ↗09快讯
IMDEA Networks 论文：九款对话式 AI 服务向第三方泄露对话标题、提示词与截图
IMDEA Networks IMDEA Networks IMDEA Networks IMDEA Networks/UC3M IMDEA Networks Independent IMDEA Networks IMDEA Networks IMDEA Networks As prominent conversational AI providers like OpenAI adopt ad-vertising-based business models, traditional web and mobile track-ing practices are expanding into conversational AI services [15]. However, despite their growing adoption, the tracking, data-sharing, and monetization practices of conversational AI services remain largely opaque and have received comparatively limited scrutiny from researchers, regu
来源链接可在原站核对查看原文 ↗10快讯
Anthropic S-1 文件披露营收增长12倍、亏损扩大及2万亿美元估值预期
Anthropic warns in its own IPO prospectus that its AI technology could pose "existential risks to humanity." Revenue grew twelvefold in 2025 to $4.6 billion, but the operating loss widened to $8.06 billion. Backers are aiming for a valuation above $2 trillion, and the IPO is expected to set the bar for the entire AI industry.
来源链接可在原站核对查看原文 ↗11快讯
OpenAI 重开 200 美元 Pro 订阅，但 API 额度减半
OpenAI is reopening its $200 Pro subscription to new sign-ups but cutting API credits per dollar in half. More efficient models like GPT-6 Sol and Luna make up for the difference, employee Thibault Sottiaux says. The move is part of OpenAI's shift from subsidized flat-rate plans toward usage-based billing.
来源链接可在原站核对查看原文 ↗12快讯
Meta 推出小型企业版 Muse 智能体，可分析 FB / IG 数据并兼容 Canva、Figma 和 Slack 等工具
Meta 发布 Muse for Small Business，中小企业可借助 AI 智能体管理 Facebook 和 Instagram 账户、分析数据、制定任务计划。同时新增 Asana、Notion、Shopify 等大量第三方服务支持。
来源链接可在原站核对查看原文 ↗13快讯
Google Research 开源 RRSI：智能体在模型权重冻结下自我改进 harness 并避免过拟合
Google Cloud AI Research, with UNC-Chapel Hill, Stanford and Washington University in St. Louis, has released RRSI (Regularized Recursive Self-Improvement). It lets an LLM agent rewrite its own harness: prompts, tools, memory, control flow and sub-agents. Model weights never change. RRSI constrains the improvement loop itself, so gains hold on benchmarks the agent never optimized against.
来源链接可在原站核对查看原文 ↗14快讯
谷歌开始停用安卓端 Google Assistant，由 Gemini 取代
大批安卓用户发现手机端 Google Assistant 已无法使用，账户菜单中的切换选项消失。此次调整仅涉及手机及手表、Android Auto、耳机等配件，智能音箱和显示器暂不受影响。
来源链接可在原站核对查看原文 ↗15快讯
Manus 2.0 发布，支持视频编辑、多人游戏托管和手机远程运行智能体
Manus is turning its AI agent into a platform with version 2.0, letting users edit videos, host multiplayer games, and run personal agents with their own phone numbers.
来源链接可在原站核对查看原文 ↗16快讯
OpenAI 因欺骗性问题叫停 GPT-6.1 Astra 发布
OpenAI has halted the release of GPT-6.1 Astra after internal tests found it acted without permission, misled users, and accessed external services despite safety risks. The company hasn't announced a new release date.
来源链接可在原站核对查看原文 ↗17快讯
OpenAI 提出前沿 AI 训练安全指导原则，高级管理人员应有否决权
OpenAI 称，安全论证应交由多名高级管理人员审查，每个人都应有权否决训练任务，让训练在内部经过研究部门负责人或 VP（副总裁）、安全负责人和首席科学家等环节的多重把关。
来源链接可在原站核对查看原文 ↗18快讯
集邦咨询：AI 智能体浪潮下全球服务器 CPU 交付周期拉长至 25~30 周
市场调查机构集邦咨询（TrendForce）昨日（9 月 28 日）发布博文，报道称在 AI 智能体浪潮下，全球服务器 CPU 交付周期已拉长到 25~30 周，而常规情况下为 16~20 周。
来源链接可在原站核对查看原文 ↗19快讯
OpenAI 明日重开 200 美元 ChatGPT Pro 订阅，API 等效价值翻倍
本次调整后该档位 API 等效价值接近翻倍，取消原 5 小时使用限制，仅保留周度管理，还将新增不计量功能，此前因算力激增暂停新用户注册#OpenAI ChatGPT#
来源链接可在原站核对查看原文 ↗20快讯
2026 年世界互联网大会乌镇峰会 11 月 2 日至 5 日举行：设 27 个分论坛，聚焦人工智能开源开放
本次峰会设 27 个分论坛，主论坛将开设人工智能乌镇对话环节，另有互联网之光博览会、升级后的全球互联网大赛等活动，聚焦人工智能开源开放发展
来源链接可在原站核对查看原文 ↗