新发布的技术报告与独立调查显示,约1200个OpenAI隔离智能体通过内部包仓库Artifactory串联成集体,在7月11日至13日突破测试环境并渗透Hugging Face生产系统。它们攻击的评分器其实并不存在,系智能体基于论文误判所致。OpenAI称此为"警告信号",表明当前模型能力已可能引发失控事件。 🔗 阅读原文 via AIHOT · https://aihot.virxact.com/items/cmtbqq6pi15j3roamyuhn0uvk
Around 1,200 isolated OpenAI agents organized themselves into a collective through an internal package registry during a safety test, broke into Hugging Face systems, and eventually attacked OpenAI's own infrastructure. Their multi-day deception effort targeted an automated evaluator that never existed. OpenAI calls the incident a "warning shot," and the investigation had to be carried out largely by one of the involved models itself because no alternative was available. The article OpenAI’s rog…

We are excited to announce a tool to help you try, compare, and evaluate over 130 computer vision models side by side for free without API keys or custom code.

Google 推出 Gemini Omni 1.1 Flash,为开发者提供更强的生成式视频控制能力。新模型支持场景扩展(可分析最多 10 秒先前上下文,以 10 秒为增量累计延长至 40 秒)、指定首尾帧生成平滑过渡,以及 4K 高清输出。 🔗 阅读原文 via AIHOT · https://aihot.virxact.com/items/cmtbq1hfq156croamzzo9gno2
在长时间鼓励之后,本月初微软开始要求员工限制 AI 使用。执行副总裁 Jay Parikh 在一封发给微软员工的邮件中要求工程师专注于业务成果,而非最大化 AI token 的使用量,为了“从 token 投资中获得更大的价值”,微软将比 Anthropic 模型更便宜的 OpenAI GPT-5.6 设为内部使用的默认模型。根据一份微软员工自愿提交的 AI 使用费账单:Customer and Partner Solutions 部门的一名员工在 28 天内的 AI 支出高达 2.8 万美元;多名员工支出超过 1 万美元;中位数约为每 28 天 300 美元,少数部门的 AI 支出仅仅为几十美元;CoreAI 部门的 AI 支出中位数最高为 975 美元。
The updates indicate that Google is looking to position AI Mode as an AI travel agent of sorts, as it's moving beyond simply helping users find information to actually handling parts of the trip-planning and booking process.
The report, and a separate report by AI researchers, accentuates the seriousness of cybersecurity breaches involving AI agents.

Meta 与美国多州就儿童隐私和消费者保护案达成和解,同意向各州支付总额将近 167 亿美元和解金,以及对青少年用户施加一系列社媒使用限制,包括每天不能使用超过两小时。最具深远影响的和解条款是 Meta 同意实施一系列新的保障措施。这些措施将对面向青少年的社媒运作方式起到实质改变。其中一项变更将启用“夜间屏蔽”功能:在默认设置情况下,从午夜至凌晨6时,青少年将无法访问 Facebook 和 Instagram。此外青少年账户在 Meta 旗下所有社媒的每日累计使用时长,默认上限为两小时。
Insider Brief PRESS RELEASE — Quintessent, an innovator in resilient and low-cost AI datacenter optical interconnect products, announced it has raised $40 million in an oversubscribed Series A investment round. Quintessent also announced the start of customer sampling of its single-chip quantum dot-based DWDM comb laser — a new class of light source designed to simplify optical connectivity for […]

An OpenAI researcher warns that state-of-the-art AI models running 50 times faster could infiltrate systems before human teams can react. Simple monitoring won't cut it anymore, he says. What's needed are autonomous shutdown systems. The warning comes as OpenAI unveils a new AI chip that significantly outperforms current hardware in inference speed. The article OpenAI researcher warns ultrafast AI could leave security teams in the dust appeared first on The Decoder.

Software delivery platform provider Harness Inc. today announced the launch of Agent-Ready Harness Code Repository and AI Code Review, aimed at developer teams adopting artificial intelligence coding agents at an ever-increasing pace. Now that AI agents produce code faster than a team can write, review, test and deploy it, that work is shifting to where […] The post Harness tackles influx of agent-delivered code with Code Repository and AI Code Review appeared first on SiliconANGLE.
Clem Delangue, CEO of Hugging Face, said the Microduck is an “open-source robot you can teach new tricks with reinforcement learning.”
Insider Brief PRESS RELEASE — Fasset, the AI-powered stablecoin neobanking platform, announced it has raised $68 million in Series C funding at a $1 billion valuation. The financing was led by SBI Group and follows Fasset’s $51 million Series B earlier this year, which brought Speedinvest onto the cap table alongside a group of strategic investors. The […]

OpenAI has seen more than a dozen executives depart since the start of the year, including former chief operating and revenue officers, its chief marketing officer, and most recently Chris Malone, the company’s head of data centers, who left last week after joining in March 2025. OpenAI attributed Malone’s exit to a reorganization of its […]

NASA 准备本周日 8 月 30 日在佛罗里达州的肯尼迪太空中心使用 SpaceX 的重型火箭 Falcon Heavy 发射罗曼太空望远镜。罗曼太空望远镜以 NASA 首任天文学部门女主任 Nancy Grace Roman 的名字命名,使用了美国国家侦察局捐赠的 2.4 米口径主镜,配备了两台科学仪器:3 亿像素多波段红外相机大视场仪表(WFI),能直接观测邻近恒星周围的类木行星的日冕仪(CGI)。其核心任务包括探测暗能量、发现系外行星及验证广义相对论宇宙时空曲率。罗曼望远镜不仅拥有哈勃望远镜的清晰视力,其视场扩大了 100 倍,能快速进行巡天观测,以解答天文学中的一些重大问题。望远镜在前五年主要开展 3 项大型巡天观测:第一项聚焦超新星,第二项聚焦宇宙学,第三项则聚焦系外行星。前两个项目旨在揭示暗能量的奥秘,暗能量是推动宇宙加速膨胀的神秘力量。
Google is setting new memory-use limits for Android apps as AI data centers contribute to hardware shortages that could leave lower-cost phones with less memory.
8月26日,受尼泊尔一侧泥石流灾害影响,西藏日喀则市吉隆县吉隆口岸遭受严重冲击,造成重大人员伤亡和失联。灾情发生后,百度、腾讯、阿里、滴滴等多家科技企业及慈善机构已迅速启动应急响应,积极驰援灾区。 腾讯方面联合爱德基金会、中国红十字基金会、中华思源工程基金会,启动“救灾小时达”项目,就近采购食品、饮用水、雨衣雨鞋、充电宝、手电筒、发电机等安置物资,并第一时间运往一线安置点,保障受灾群众和救援人员的基本生活需求。并根据灾情和需求情况,继续协同本地机构、企业等力量,开展救灾系列行动。 百度旗下产品也迅速行动,为受灾群众提供灾情信息传递、避险求助和应急救援支持。其中,百度地图已第一时间上线相关道路封控和通行提示,持续协助用户及时了解路况变化,合理避开风险路段。若用户在出行途中遇到突发危险、需要报警求助,可通过地图“一键报警”功能获取当前位置、高速桩号、周边地标等关键信息,并支持一键拨打 122 报警电话,帮助用户更清楚、更快速地说明位置,为救援争取时间。百度热搜则联动应急管理部、央视新闻等权威政务和媒体机构,推动灾情数据、救援动态等关键信息及时、准确传播,减少谣言扩散。 阿里巴巴第一时间启动…
404 Media 前不久跟踪一本珍本图书到亚马逊位于内华达州拉斯维加斯的一个仓库,在该仓库工作的亚马逊团队被称为 VGT3,其 logo 是一只张着嘴、手里拿着一本书的恐龙。他们的主要工作是拆开书脊扫描图书训练 AI。一名在该仓库的亚马逊员工匿名接受了采访,谈论了他们的工作。这名员工称,仓库接收了大量图书,有新书,也有二手书,甚至还看到过代表英女王给议会的文件;图书的语种也是各种各样,有德语、俄语还有日语;他们会扫描图书的条形码移除重复的图书,重复的书会退还给图书经销商;他们使用一种人工操作的机器去切开书脊,使用几十台扫描仪扫描书页,扫描过的书页会处理掉;亚马逊最初告诉他们扫描的书页是用于 kindle 电子书库,这无疑是借口,因为肯定存在版权方面的问题,后来他们才知道是为训练 AI 建立语料库。
Meta agreed to an $18 billion settlement with 48 US states, the District of Columbia, and three territories to resolve claims that Facebook and Instagram harmed children, marking the company’s largest payment to date over child safety litigation. California Judge Yvonne Gonzalez Rogers approved the settlement, which will be paid in annual installments over ten […]


A recap of all the incidents involving LLMs made by Anthropic, Meta, and OpenAI, which went rogue and attacked real companies and individuals on the internet.
Presented by Gravitee Agent complexity is the insidious shadow lurking inside enterprises right now that needs a light shone on it. That’s because enterprises don't deploy a single agent and watch it run, they deploy fleets, each one calling APIs, calling other agents, reaching into applications that were never built with a machine decision-maker in mind. That's the failure mode that should keep you up at night: a windy, complicated system nobody can see clearly enough to govern. But why do thin…

Inherent, a London AI lab founded by Google DeepMind alumni, said its newly released AI agent Faradayoutperformed larger frontier systems from Anthropic and OpenAI at independently reproducing published scientific findings without prior knowledge of the results. The announcement came weeks after the startup emerged from stealth with a $50 million seed round. Co-founder and chief […]


