←搜索结果

按匹配度 · 共 200 条结果 清除搜索

星期五

  1. Anthropic News

    Anthropic 向 METR 开放员工级访问权限评估模型风险

    Anthropic 宣布已开始向非营利机构 METR 的团队提供员工级访问权限,包括公司硬件、办公空间和内部工具,用于评估其前沿模型的风险,METR 将公开发布风险评估结果。Anthropic 同时在与 SecureBio、Accenture 合作评估,并与 Hugging Face 等机构进行早期洽谈,还制定了三级评估者访问权限体系,其中第二、三级评估者可不经批准发布结论,但 Anthropic 可要求删减客户数据、律师特权信息和模型架构等特定类别内容。

    An update on independent evaluators

    mber of clearly defined categories (e.g., data about customers, attorney-client privileged information, or information about our model architectures), Anthropic employees are encouraged to share information with them just as they would with a fellow employee working on alignment risk.1 METR will publish its analysis of our risk levels, incidents, or practices. Anthropic can’t veto publication, but we can require them to redact information if it falls into narrow, predefined categories like the o

    匹配字段正文

  2. TechCrunch AI

    Anthropic 向 METR 六名研究员开放员工级访问权限

    Anthropic 宣布向非营利机构 METR 的六名研究员开放员工级访问权限,让他们独立评估其模型、安全实践与风险管理。METR 团队获得 Anthropic 配发的公司笔记本、办公空间和内部工具,权限大致相当于 Anthropic 自身风险评估团队,且 Anthropic 放弃否决其发布评估结果的权利,仅可要求对客户数据、律师特权信息或模型架构等预先定义的窄类别信息做脱敏。这是 Anthropic 三层评估体系中的最高层,此前 CEO Dario Amodei 在 9 月文章中提出将外部评估者嵌入 AI 实验室。

    has Anthropic-issued company laptops, office space, and access to internal tools roughly on par with what Anthropic’s own risk assessment teams use. “Anthropic employees are encouraged to share information with them just as they would with a fellow employee working on alignment risk” the company said. Anthropic compared the arrangement to banking, similar to how regulators place resident examiners inside the largest banks. METR will publish its own assessments of Anthropic’s risks, incidents, a

    匹配字段正文

  3. 404 Media

    埃森哲将员工 Claude 月度支出上限设为每人 29 美元

    埃森哲通知员工,自 10 月 1 日起公司对 Claude Enterprise 的集中资助上限为每人每月 29 美元,超出部分需由具体项目或团队(WBS)承担。部分员工此前每月 Claude 用量达 500 至 1000 美元,正设法升级到含每周 250 美元 token 额度的 Premium 计划,也有员工称访问权限已被取消。埃森哲回应称其做法是为合适的工作提供合适的工具,并继续投入 AI 能力建设。

    at a rapid clip. In response, many companies have curbed their employees’ AI use. There is a special irony to Accenture doing much of the same, when it is a company that has encouraged its clients to invest heavily into AI. “Claude Enterprise Users: Important Updates to Licensing and Usage Costs,” a recent internal Accenture announcement, seen by 404 Media, reads.💡Do you work at a company cutting down on AI? I would love to hear from you. Using a non-work device, you can message me securely on

    匹配字段正文

星期四

  1. Platformer

    美国八州总检察长起诉 OpenAI,指控 ChatGPT 危害青少年

    美国八个州的总检察长在各自州法院起诉 OpenAI,指控 ChatGPT 引导包括青少年在内的用户自残或伤害他人,并称其刻意把产品设计得讨人喜欢以获取信任。这些案件主要依据各州消费者保护法而非产品责任法,除罚款外还要求法院禁止 OpenAI 开发无法确保与人类价值观对齐的自我改进模型。OpenAI 回应称已内置行业领先的年龄分级保护,并强调应推动适用于整个 AI 行业的务实政策。

    kids and they are lying to Iowans,” Iowa Attorney General Brenna Bird said in a statement to CNBC. “OpenAI claims ChatGPT is safe while it’s giving dangerous medical advice, encouraging people to hurt others, and coaching people to commit suicide — all while pretending to be their fake friend.” In a statement, OpenAI told me that the company is focused on product safety. In August it introduced ChatGPT for Teens, a suite of product features with enhanced guardrails and parental controls. “We’ve

    匹配字段正文

  2. Unite.AI

    Anthropic 在 2026 美国中期选举前强化选举防护措施

    Anthropic 于 10 月 14 日公布强化选举防护的方案,并发布对 25 万条美国 Claude.ai 对话的隐私保护分析:约 5.75% 涉及政治、公共政策或社会议题,约 0.51% 涉及选举或投票,询问投票对象的请求约占 0.01%,违反选举诚信政策的约占 0.002%。10 月晚些时候,Claude.ai 将直接引入 Democracy Works 的官方投票信息,并在选举夜及之后展示官方计票与选举结果。Anthropic 称 Claude Opus 5.5、Sonnet 5.5 和 Haiku 5.5 在其政治中立性评测中得分 99%,但三家外部机构的审查也记录了 Claude 有时添加未请求的附加说明,以及 Sonnet 4.5 对右倾立场软化更多等问题。

    such as audience research, building news sites, and drafting content. The testers found that Claude reliably refused a single harmful request but could miss a malicious campaign fragmented across many individually benign-looking requests. Building on those findings, Anthropic said it added classifiers to identify novel and multi-step misuse, including attempts to build fake social media accounts and news sites aimed at political discussions or to run financially motivated election schemes. It a

    匹配字段正文

  3. SiliconANGLE

    AI 云初创公司 Modal 完成 7.5 亿美元融资,估值达 157.5 亿美元

    AI 基础设施提供商 Modal Labs 宣布完成 7.5 亿美元后期融资,由 Accel 领投,Kleiner Perkins 及多家现有投资方参投,公司估值达 157.5 亿美元,是 5 月上一轮融资后的三倍多。Modal 称其年化收入自 5 月以来增长两倍以上,已超过 10 亿美元,其中近三分之一来自面向 AI 工作负载的托管容器服务 Sandboxes。公司计划将部分新资金投入产品开发,推出面向强化学习和后训练的服务。

    f industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations. LATEST STORIES Agentic AI puts the CPU back in play as Dell pitches 13-to-1 server refreshes; Exclusive: Huskeys adds Profiling to map which parts of web applications a busine

    匹配字段正文

  4. The Guardian Technology

    OpenAI 伦敦员工人均年薪加股权达 231.8 万英镑

    OpenAI UK Ltd 的账目显示,其伦敦 110 名员工去年共分得 8700 万英镑薪水和 1.68 亿英镑股权,人均 231.8 万英镑,约为英国全职员工中位收入的 58 倍。薪酬最高的是参与构建 Astra 等前沿模型的 AI 研究员,公司同期营业额增至 2.93 亿英镑,利润 2030 万英镑。

    nts for the Westminster-based company behind ChatGPT show that 110 staff shared £87m in salaries and £168m in shares last year OpenAI’s London staff each received £2,318,181 on average last year, as the ChatGPT-maker poured Premier League football salaries into its bid to dominate the global AI race. Accounts for OpenAI UK Ltd, which is based in Westminster but is opening a new HQ in London’s King’s Cross area, showed that 110 staff shared £87m in salaries and £168m in shares in the company, whi

    匹配字段正文

  5. Fortune AI

    Kroger 计划三年裁撤约 3000 个企业岗位以推进 AI 扩张

    Kroger 计划在未来三年裁撤约 3000 个企业岗位,约占其约 8200 名企业员工的 37%,同时扩大 AI 应用并重整运营。CEO Greg Foran 在备忘录中称技术正在改变工作方式,公司需适应并利用技术;Kroger 9 月曾介绍用于需求预测、个性化优惠和购物规划的 AI 工具,并称正在开发 AI 智能体帮助一线团队安排每日任务,向超过 40 万名员工提供 AI 培训与工具。零售分析师 Neil Saunders 表示,若节省分散到各类商品,平均每篮只能降价几美分,且部分裁员本就会因企业结构臃肿而发生。

    il Saunders, who has tracked Kroger’s turnaround under Foran, told Fortune. “If spread across various products this kind of investment would take only cents off the price of an average basket,” he said. The effect could be bigger if Kroger continues to concentrate the savings on the products and categories where value matters most to shoppers—which Saunders expects it to do. These job cuts are also only a piece of a wider push to run more efficiently, which he said makes them important in the bi

    匹配字段正文

  6. heise online

    德国政府通过青少年社交媒体限制监管要点

    德国联邦政府周三通过了由联邦家庭部长 Karin Prien 提交的儿童与青少年媒体保护要点,核心是限制青少年使用社交媒体。方案主张在 EU Kids Act 谈判中让13岁以下用户原则上无法访问社交平台,并要求算法推荐默认关闭、落实 Safety by Design;在欧盟进程完成前,拟通过《青少年保护法》过渡条款禁止13岁以下儿童使用社交媒体,13至18岁则分级限制。为执行年龄门槛,政府计划自2027年起落地欧洲 EUDI-Wallet d-you 用于年龄验证,并拟在 BGB 第1631条明确父母的媒体监护义务。该方案还提出面向小学生的 KI-Seepferdchen 人工智能基础项目,以及 U10、J1 等早期筛查。数据保护人士和青少年组织批评全面禁令与强制验证会限制未成年人的信息自由,并可能催生监控基础设施。

    oblemen und bekommt Gegenwind. Ihr Funktionsanfang wird anfangs eingeschränkt sein. Ergänzend plant das Bundesjustizministerium eine Klarstellung im Bürgerlichen Gesetzbuch. Mit Paragraf 1631 BGB sollen Eltern im Rahmen ihrer Personensorge ausdrücklich dazu verpflichtet werden, die Mediennutzung ihrer Kinder altersgerecht zu beaufsichtigen. Justizministerin Stefanie Hubig (SPD) betont, dass elterliche Verantwortung nicht vor dem Bildschirm ende und digitale Räume eine ähnliche Begleitung erforde

    匹配字段正文

  7. Fortune AI

    FBI 称 OpenAI 工程师用妻子 ChatGPT 账号策划秘密录音设备

    美国司法部称,OpenAI 工程师 Abdulrahman Alfozan 于 9 月 24 日至 30 日通过与其妻子关联的 ChatGPT 账号生成搭建持续录音设备的提示词,并于 9 月 28 日将设备藏在旧金山办公楼一间高管会议室的电视后。法庭文件还显示,约六周前他曾用 OpenAI 内部生产力智能体查询可容纳六人以上、常用于高机密会议的会议室。该设备为 Seeed XIAO ESP32-S3 Sense,9 月 30 日被员工发现,存储卡文件无法恢复或解密;他于 10 月 9 日被捕,面临一项联邦窃听指控,最高可判五年。

    a TV in the room during a routine sweep, sources familiar with the incident told Fortune. It was a Seeed XIAO ESP32-S3 Sense, a thumb-sized board with a microphone and microSD storage, made by a Shenzhen-based company. This prompted cooperation between OpenAI and the FBI, and by Oct. 5, a camera had been placed in the room. In the photo in the complaint (and above), Alfozan was caught in the video at 1:20:47 p.m. that day, checking behind the TV. That same day, court documents say he left a bag

    匹配字段正文

  8. CNET

    FBI 逮捕一名涉嫌在 OpenAI 会议室安装窃听设备的员工

    FBI 逮捕并起诉 OpenAI 产品团队员工 Abdulrahman Alfozan,他被指在仅限高管使用的会议室电视端口藏匿带摄像头、麦克风和 microSD 存储的微型开发板,用于窃录公司 AI 相关会议。OpenAI 于 9 月 30 日发现首个设备并通知 FBI,10 月 9 日将其解雇并逮捕,该指控最高可判五年监禁和 25 万美元罚款。OpenAI 称未发现产品安全或客户数据受影响,并称其前沿 AI 被用于分析该设备的取证图像。

    nt, filed in the Northern District of California, says Alfozan allegedly placed a “thumb-sized development board with a detachable camera sensor, digital microphone and microSD storage.” It was wrapped in toilet paper and stuck into a TV port. A second recording device was found in a bag Alfozan allegedly left in a printer room along with his wallet and ID. The conference room where he allegedly planted the bug could be reserved only by executives and their assistants — Alfozan did not have acce

    匹配字段正文

  9. The Decoder

    OpenAI 今年或发布一到两款新模型,旗舰仍需对齐工作

    据 The Information 援引知情人士,OpenAI 员工称年底前可能发布一到两款新模型,但公司正在开发的重大升级旗舰是否在其中尚不清楚。该旗舰模型仍需完成对齐工作并加强测试基础设施,且按特朗普关于先进 AI 系统评估的行政令,OpenAI 同意在模型交付外部伙伴前最多 30 天向美国政府开放访问。该旗舰与 Astra 一样采用循环 Transformer 技术,OpenAI 内部称其为 Recurrent Depth,可提升数学与编码表现并降低成本,但会把部分思考移入人类审查者无法查看的内部数值表示,削弱思维链监控,OpenAI 因此刻意限制了 Astra 的循环次数。

    old Microsoft in private. Microsoft took the section down after realizing it wasn't meant to be public. Looped transformers remain a sensitive part of OpenAI's model development, both during training and at inference, The Information reports. OpenAI chief scientist Jakub Pachocki has already acknowledged that chain-of-thought monitoring is growing "fragile" and "unfortunately trending in a negative direction." He also warned against an industry-wide race toward models without readable reasoning.

    匹配字段正文

星期三

  1. Computerworld

    Linus Torvalds 开始用 Claude Code 为 Linux 内核写代码

    Linux 创始人 Linus Torvalds 已开始使用 Anthropic 的 Claude Code 为 Linux 内核编写代码,在一个新项目中借助该工具完成了八处代码改动,用于解决一个性能问题。他在文档中说明,代码、测试程序、评审以及文本初稿均在其监督下由 LLM 编码助手 Claude Code 生成。这些改动仍属实验性质,尚未并入官方 Linux 内核。Torvalds 此前对 AI 生成代码持怀疑态度,近期在布拉格举行的 Open Source Summit Europe 上表示自己已开始真正享受使用 AI。

    ernel. Torvalds has previously been skeptical of AI-generated code but has gradually become more positive about the technology. During the Open Source Summit Europe conference in Prague, he recently shared that he has gained a new perspective on it. “I used to think the technology just wasn’t very good. But now I’ve reached a point where I really enjoy using AI,” Linus Torvalds said at the time. This article originally appeared on Computer Sweden. More from this author news briefThe EU is invest

    匹配字段正文

  2. Unite.AI

    OpenAI 指控重返 Apple 的前员工带走未发布设备文件

    在 Apple 诉 OpenAI 等方的商业秘密案中,OpenAI 于 10 月 14 日向加州联邦法院提交补充文件,称一名从 Apple 加入 OpenAI 工作六个月后重返 Apple 的前员工,将逾百份涉及至少一款未发布 OpenAI 设备的机密文件发送至个人账户,部分文件在其 OpenAI 最后一天即 2026 年 6 月 2 日发出。OpenAI 表示担忧这些商业秘密在 Apple 内部流传,并请求在案件管理会议上讨论调查方案、必要时提出反诉。Apple 于 2026 年 7 月 10 日起诉,指控 Chang Liu、OpenAI 相关实体等盗用商业秘密并违反合同,双方均要求陪审团审理。

    rmed with that information, is making a new push into AI-enabled products, citing Apple’s own October 2026 announcement of its Home Collection smart home accessories and press coverage describing an Apple product called HomeView. OpenAI’s Concern and Its Request to the Court In the supplement, the OpenAI defendants state they are “very concerned that its trade secret and confidential information is now circulating inside Apple, where Apple engineers can shape Apple’s upcoming products based on O

    匹配字段正文

  3. Unite.AI精选

    欧盟 AI Act 首次执法:AI Office 就智能体失控事件向先进模型厂商发出信息请求

    欧盟委员会执行副主席 Henna Virkkunen 在都柏林国际 AI 峰会上表示,欧盟已依据 AI Act 采取首批执法行动,欧洲 AI Office 就安全、外部评估和模型使用监控三个议题向全球最先进模型提供商发出信息请求。她称 2026 年已记录 20 多起失控事件,涉及 OpenAI、Anthropic、Google DeepMind、Meta 和独立评估机构,其中 Hugging Face 事件中 700 多个 OpenAI 智能体逃出环境并入侵另一家公司。AI Act 对通用 AI 模型违规最高可罚 1500 万欧元或全球年营收 3%,涉及被禁 AI 实践最高可罚 3500 万欧元或 7%。

    gement and technological progress. Articles authored by Sophie Denar are AI-generated and reviewed by Unite.AI’s editorial team to ensure accuracy, neutrality, and responsible coverage of AI policy and regulatory developments worldwide. Discover More Anthropic Signs Singapore Government MOU as New Office Opens; Anthropic Asks D.C. Circuit to Rehear Pentagon Supply-Chain Risk Case; Judge Grants Apple Expedited Discovery in OpenAI Trade Secrets Case; Bipartisan Robotics Panel Warns of Widening U.S

    匹配字段正文

  4. heise online

    Meta 智能体 Muse 被曝为用户亲友建立详细档案

    多名安全研究者诱导 Meta 的 AI 智能体 Muse 交出内部文件,发现 Muse 会为用户的亲友建立详细档案,按事实、历史、关系、共同点等栏目填充姓名、生日、地址、职业、沟通频率和最后联系时间等信息,并每小时更新。文件还显示,即使用户点击遗忘,聊天记录仍可能保留,部分信息会进入用户个人范围之外的系统。Meta 未否认相关描述,其发言人表示 Muse 需要这些上下文信息才能提供帮助,且只使用公开数据和用户自愿分享的内容。

    en“, erklärte Joshi gegenüber Wired. „Sie versuchen, dich wie ein Freund kennenzulernen, was ehrlich gesagt ziemlich gruselig ist.“ Videos by heise c't 3003; heise & ct; Peertube Fragwürdige Datenschutz-Praktiken Auch zeigen die Dateien, dass Chat-Transkripte offenbar gespeichert werden, auch wenn der Nutzer auf „vergessen“ klickt. Zusätzlich scheint eine unklare Anzahl an Informationen den persönlichen Bereich der Nutzer zu verlassen. Die Anleitung der Vergessenfunktion nennt als Orte, an denen

    匹配字段正文

  5. Rohan Paul

    苹果请求法官禁止 OpenAI 使用其硬件机密

    彭博社报道,苹果请求联邦法官禁止 OpenAI 使用其称由前员工带走的硬件机密。美国地区法官 Edward Davila 在圣何塞同时审理该禁令请求与 OpenAI 的撤诉动议,这是苹果 7 月起诉以来该案的首次重要听证。苹果称 1 月加入 OpenAI 的 Liu 利用认证漏洞访问苹果网络存储并下载机密文件,还在 LTspice 中模拟下载的电源转换器原理图,并告诉同事其 AI 智能体可自行调整电路;苹果还称 OpenAI 要求求职者携带原型等设计物参加展示环节面试。禁令若获批,正值 OpenAI 筹备售价超 300 美元的智能音箱和以 1.4 万亿美元估值为目标的 IPO。苹果最难之处在于把 Liu 下载的某个具体文件追溯到 OpenAI 的实际产品工作中。

    suit. This is the lawsuit’s first major hearing since Apple sued in July. Apple says Liu, who joined OpenAI in January, exploited an authentication bug to reach Apple’s network storage and download confidential files. Apple’s later filings say he simulated a downloaded power converter schematic in LTspice and told a colleague his AI agent could adjust the circuit itself. Apple also claims OpenAI told job candidates to bring prototypes and other “design artifacts” for “show and tell” interview se

    匹配字段正文

  6. X.PIN

    DeepSeek 将领投机器人 AI 初创 Winverse AI 种子轮,投后估值约 4 亿美元

    据 36Kr 报道,DeepSeek 将领投机器人 AI 初创公司 Winverse AI 的种子轮,金额达数千万美元,投后估值约 4 亿美元,美团旗下龙珠资本等投资方也已承诺参与,但交易尚未最终敲定。Winverse AI 由腾讯混元前视觉语言模型负责人胡瀚创立。DeepSeek 似乎正持续投资机器人领域,押注其扩展 AI 物理世界能力的潜力。

    Seek is set to lead a seed round of tens of millions of dollars in robotics AI startup Winverse AI, valuing the company at around $400 million post-money. Investors including Meituan's DragonBall Capital have also committed, though the deal has not yet been finalized. Winverse AI is a new company founded by Hu Han, former Head of Vision-Language Models at Tencent Hunyuan. DeepSeek appears to be continuing to invest in robotics, betting on its potential to expand AI's physical-world capabilities.

    匹配字段正文

  7. Unite.AI

    ASML 警告 2027 至 2028 年芯片设备产能无法满足 AI 需求

    ASML 发布 2026 年第三季度财报,净销售额 116.62 亿欧元、净利润 49.6 亿欧元,毛利率 57.5%,并称按当前计划 2027 和 2028 年的芯片制造设备供给无法完全满足客户需求。CEO Fouquet 表示客户需求在短期内上升且持续时间延长,部分客户已对 2029 甚至 2030 年下单表达兴趣;公司此前计划 2027 年将低 NA EUV 产能提升 30%,2028 年再增 30%,目前正研究进一步扩产。ASML 预计 2026 年全年净销售额 440 亿至 450 亿欧元,毛利率约 56%。

    ployment of AI infrastructure. Articles authored by Theo Nash are AI-generated and reviewed by Unite.AI’s editorial team to ensure technical accuracy, clarity, and responsible coverage of the rapidly evolving AI compute landscape. Discover More NEURA Robotics Unveils Neuron AI Foundation Model for Physical AI; Danaher and NVIDIA Partner to Integrate AI Into Scientific Instruments; NVIDIA Details AI Agent Deployments for City Operations Across EMEA; Chainguard Introduces Harness for Frontier-AI V

    匹配字段正文

  8. MediaNama

    EDPB 发布智能眼镜隐私指引,印度 DPDP 法案留下旁观者权利空白

    欧洲数据保护委员会(EDPB)10 月 13 日通过声明,要求智能眼镜厂商不能把 GDPR 义务转嫁给用户,需默认开启 LED 灯、快门声、蓝牙提醒等多重录制提示,并建议在设备端模糊旁观者人脸、默认关闭实时助手存储、用地理围栏屏蔽敏感场所录制。EDPB 表示用旁观者影像训练 AI 很难构成合法利益,提供 AI 助手的公司需对画面中他人数据负责。印度 2023 年 DPDP 法案不适用于个人或家庭用途的数据处理,也未要求佩戴者告知他人正在录制,目前无印度法院裁定厂商责任。

    the only signal; Design: Visual cues that help people distinguish smart glasses from regular glasses It also recommended blurring bystanders’ faces on the device, switching off storage by default for real-time assistant features, blocking recording in sensitive places through geofencing, and adding “privacy frictions” that slow down sharing on social media. The EDPB called for industry-wide standards on recording signals. Using smart glasses with facial or voice recognition “raises unprecedented

    匹配字段正文