霍尔木兹海峡关闭后 亚洲经济体依托多元采购与储备避免严重能源短缺
霍尔木兹海峡关闭导致全球市场每日原油供应骤减约1000万桶。七个月过去后,亚洲各国基本避免了严重的能源短缺局面。面对供应危机,亚洲原油进口国积极采取应对措施:利用沙特阿拉伯和阿联酋的输油管道改道运输,动用战略储备,并在美国制裁豁免下扩大了对美、俄两国的原油采购。其中,中国通过动用储备和调节需求减少了原油进口,而美国创纪录的液化天然气(LNG)发运量则弥补了卡塔尔供应中断造成的损失。尽管避免了最灾难性的局面,该地区仍承受着持续的经济压力:目前国际原油价格仍徘徊在每桶100美元左右,液化天然气现货价格翻了一倍以上;随着冬季临近,巴基斯坦、孟加拉国等经济较为薄弱的国家已面临轮流停电与限电配给的严峻考验。
—— The New York Times 2026-09-29 UTC
Asian Economies Avert Severe Energy Shortages Following Strait of Hormuz Closure Through Diversification and Reserves
霍尔木兹海峡关闭导致全球市场每日原油供应骤减约1000万桶。七个月过去后,亚洲各国基本避免了严重的能源短缺局面。面对供应危机,亚洲原油进口国积极采取应对措施:利用沙特阿拉伯和阿联酋的输油管道改道运输,动用战略储备,并在美国制裁豁免下扩大了对美、俄两国的原油采购。其中,中国通过动用储备和调节需求减少了原油进口,而美国创纪录的液化天然气(LNG)发运量则弥补了卡塔尔供应中断造成的损失。尽管避免了最灾难性的局面,该地区仍承受着持续的经济压力:目前国际原油价格仍徘徊在每桶100美元左右,液化天然气现货价格翻了一倍以上;随着冬季临近,巴基斯坦、孟加拉国等经济较为薄弱的国家已面临轮流停电与限电配给的严峻考验。
Seven months after the closure of the Strait of Hormuz removed approximately 10 million barrels of crude oil per day from global markets, Asian nations have largely avoided acute energy shortages. Importers adapted by rerouting crude through Saudi Arabian and Emirati pipelines, tapping strategic stockpiles, and purchasing expanded shipments from the United States and Russia under U.S. sanctions waivers. China reduced crude imports by drawing down reserves and moderating demand, while record American liquefied natural gas shipments offset Qatari supply losses. Despite avoiding catastrophe, the region faces sustained economic strain: crude remains around $100 per barrel, spot LNG prices have more than doubled, and poorer nations such as Pakistan and Bangladesh face rolling blackouts and power rationing ahead of winter.
—— The New York Times 2026-09-29 UTC
受美国制裁事件触动 荷兰开发基于NixOS的政务系统以替代微软
荷兰政府已启动一项名为“政府数字化自主工作环境”(DAWO)的项目,计划将公共部门的信息基础设施从微软(Microsoft)软件转向基于NixOS构建的开源生态系统。促成这一决定的直接导火索,是美国此前对设在海牙的国际刑事法院实施制裁,导致该法院首席检察官的微软服务访问权限被突然切断,从而暴露出依赖外国软件的严重风险。该项目由荷兰内政与王国关系部负责监管,并由政府IT机构SSC-ICT、DICTU与DUO-ICT联合开发。目前,荷兰已有8个市镇展开试点运行,首个面向全荷兰政府系统的稳定版本预计将于2027年底正式推出。
—— Tom's Hardware 2026-09-28 UTC
Netherlands Develops NixOS-Based Government System to Replace Microsoft Following US Sanctions
荷兰政府已启动一项名为“政府数字化自主工作环境”(DAWO)的项目,计划将公共部门的信息基础设施从微软(Microsoft)软件转向基于NixOS构建的开源生态系统。促成这一决定的直接导火索,是美国此前对设在海牙的国际刑事法院实施制裁,导致该法院首席检察官的微软服务访问权限被突然切断,从而暴露出依赖外国软件的严重风险。该项目由荷兰内政与王国关系部负责监管,并由政府IT机构SSC-ICT、DICTU与DUO-ICT联合开发。目前,荷兰已有8个市镇展开试点运行,首个面向全荷兰政府系统的稳定版本预计将于2027年底正式推出。
The Dutch government has launched an initiative called DAWO (Digital Autonomous Work Environment for Government) to transition its public sector infrastructure from Microsoft software to an open-source ecosystem built on NixOS. The move was prompted by U.S. sanctions against the International Criminal Court in The Hague, which abruptly cut off the chief prosecutor's access to Microsoft services and highlighted the risks of foreign software dependency. Overseen by the Ministry of the Interior and Kingdom Relations, the project is being developed by SSC-ICT, DICTU, and DUO-ICT. Eight Dutch municipalities are currently conducting trial runs, with the first stable government-wide release scheduled for the end of 2027.
—— Tom's Hardware 2026-09-28 UTC
德里推进电网现代化与配电私有化 配电损耗从50%降至5%
2002年至2026年间,德里将配电网损耗率从50%以上大幅降至5%至6%,电网供电可靠性也从约70%跃升至99.9%以上。
2002年德里电力局(Delhi Vidyut Board)重组拆分,随后印度联邦《2003年电力法》出台,私营配电企业塔塔电力(Tata Power)与BSES对当地电网展开了全面改造。为遏制技术性损耗与窃电现象,改造工程整合了数据采集与监视控制系统(SCADA)、用于稳定无功功率的电容器组、架空绝缘电缆以及数字化电表。此外,电力公用事业公司还结合社区宣传与人工智能技术追查窃电并强化电费收缴,确保了电网能够稳定支撑起8748兆瓦的历史最高峰值负荷需求。
—— IEEE Spectrum 2026-09-28 UTC
Delhi Slashes Power Distribution Losses from 50% to 5% Through Grid Modernization and Utility Privatization
2002年至2026年间,德里将配电网损耗率从50%以上大幅降至5%至6%,电网供电可靠性也从约70%跃升至99.9%以上。
2002年德里电力局(Delhi Vidyut Board)重组拆分,随后印度联邦《2003年电力法》出台,私营配电企业塔塔电力(Tata Power)与BSES对当地电网展开了全面改造。为遏制技术性损耗与窃电现象,改造工程整合了数据采集与监视控制系统(SCADA)、用于稳定无功功率的电容器组、架空绝缘电缆以及数字化电表。此外,电力公用事业公司还结合社区宣传与人工智能技术追查窃电并强化电费收缴,确保了电网能够稳定支撑起8748兆瓦的历史最高峰值负荷需求。
Delhi reduced its power distribution losses from over 50 percent in 2002 to between 5 and 6 percent by 2026, boosting grid reliability from roughly 70 percent to over 99.9 percent. Following the unbundling of the Delhi Vidyut Board in 2002 and the federal Electricity Act of 2003, private distributors Tata Power and BSES overhauled the network. The turnaround integrated supervisory control and data acquisition (SCADA) systems, capacitor banks for reactive power stabilization, insulated overhead cables, and digital metering to curb technical losses and electricity theft. Utilities also utilized community outreach and artificial intelligence to trace theft and enforce collections, allowing the grid to reliably support an all-time peak demand of 8,748 megawatts.
—— IEEE Spectrum 2026-09-28 UTC
麦当劳在全美近1.4万家餐厅部署AI定价系统 引发加盟商分歧与监管担忧
麦当劳正在其全美近1.4万家餐厅部署一套人工智能定价引擎,依据消费者的支付意愿及当地竞品数据,为各门店量身定制最佳价格。路透社的一项调查显示,该平台由麦当劳与泰格分析公司(Tiger Analytics)联合开发,导致相邻门店之间的价差进一步拉大。尽管麦当劳坚称加盟商可自愿使用该工具,但不少加盟商表示面临来自总部的巨大施压,要求其执行系统建议价格,双方由此在营业利润率问题上产生摩擦。麦当劳希望通过这一策略以量取胜、拉动公司营收,但该模式既面临消费者反弹的风险,也因涉嫌潜在的算法协同定价而招致更严格的反垄断审查。
—— Reuters 2026-09-29 UTC
McDonald's Deploys AI Pricing Engine Across Nearly 14,000 U.S. Restaurants Amid Franchisee and Regulatory Tensions
麦当劳正在其全美近1.4万家餐厅部署一套人工智能定价引擎,依据消费者的支付意愿及当地竞品数据,为各门店量身定制最佳价格。路透社的一项调查显示,该平台由麦当劳与泰格分析公司(Tiger Analytics)联合开发,导致相邻门店之间的价差进一步拉大。尽管麦当劳坚称加盟商可自愿使用该工具,但不少加盟商表示面临来自总部的巨大施压,要求其执行系统建议价格,双方由此在营业利润率问题上产生摩擦。麦当劳希望通过这一策略以量取胜、拉动公司营收,但该模式既面临消费者反弹的风险,也因涉嫌潜在的算法协同定价而招致更严格的反垄断审查。
McDonald's is deploying an artificial intelligence pricing engine across its nearly 14,000 U.S. restaurants to generate location-specific optimal prices based on customer willingness to pay and local competitor data. An investigation by Reuters reveals that the platform, developed with Tiger Analytics, has widened price discrepancies between nearby stores. While McDonald's maintains the tool is non-mandatory, franchisees report significant corporate pressure to adopt recommended prices, prompting friction over operating margins. The strategy aims to increase corporate revenue via sales volume, but it faces risks of consumer backlash and heightened antitrust scrutiny over potential algorithmic price coordination.
—— Reuters 2026-09-29 UTC
贸易争端升级 美国正式对近10亿美元加拿大商品实施进口禁令
2026年9月29日,美国针对价值约9.67亿美元加拿大进口商品的禁令正式生效,重点涉及酒类、乳制品及摩托车。今年入夏之初,美方曾对约200亿美元的加拿大商品加征50%的关税,引发加拿大随后出台等额反制关税;此次禁令的实施标志着两国持续发酵的贸易争端进一步升级。
分析人士指出,此次受波及的商品(包括庞巴迪动力产品公司(Bombardier Recreational Products)生产的三轮摩托车以及乳清等)在两国每年高达8800亿美元的双边贸易额中仅占极小一部分。然而,该举措不仅加剧了政治层面的紧张局势,为未来《美墨加协定》(USMCA)的续约前景蒙上阴影,还将促使加拿大总理马克·卡尼(Mark Carney)加快推进出口市场多元化战略,进一步拓展欧盟、印度及中国市场。
—— AP News 2026-09-29 UTC
US Enacts Ban on Nearly $1 Billion in Canadian Imports Amid Escalating Trade Dispute
2026年9月29日,美国针对价值约9.67亿美元加拿大进口商品的禁令正式生效,重点涉及酒类、乳制品及摩托车。今年入夏之初,美方曾对约200亿美元的加拿大商品加征50%的关税,引发加拿大随后出台等额反制关税;此次禁令的实施标志着两国持续发酵的贸易争端进一步升级。
分析人士指出,此次受波及的商品(包括庞巴迪动力产品公司(Bombardier Recreational Products)生产的三轮摩托车以及乳清等)在两国每年高达8800亿美元的双边贸易额中仅占极小一部分。然而,该举措不仅加剧了政治层面的紧张局势,为未来《美墨加协定》(USMCA)的续约前景蒙上阴影,还将促使加拿大总理马克·卡尼(Mark Carney)加快推进出口市场多元化战略,进一步拓展欧盟、印度及中国市场。
On September 29, 2026, a United States ban on approximately $967 million worth of Canadian imports officially took effect, targeting alcoholic beverages, dairy products, and motorcycles. The policy escalates an ongoing trade dispute following 50% U.S. tariffs placed earlier in the summer on roughly $20 billion in Canadian goods, which prompted dollar-for-dollar retaliatory tariffs from Canada. Analysts note the targeted goods, including Bombardier Recreational Products' three-wheel motorcycles and whey, constitute a minor fraction of the $880 billion annual bilateral trade flow. However, the measure exacerbates political tensions, clouds the future renewal of the US-Mexico-Canada Agreement (USMCA), and accelerates Canadian Prime Minister Mark Carney's initiatives to diversify export markets toward the European Union, India, and China.
—— AP News 2026-09-29 UTC
IPO文件显示Anthropic拟投入逾5180亿美元建设AI基础设施 多数协议不可撤销
路透社查阅的一份机密IPO招股书显示,人工智能初创公司Anthropic预计将在未来10年内,通过六家合作伙伴在AI基础设施上投入至少5180亿美元。这一建设计划堪称有记录以来规模最大的算力采购承诺之一。文件指出,上述财务承诺中约80%为不可撤销协议,或属于无论实际使用情况如何均须全额支付的款项。此外,招股书还披露了其与xAI及超威半导体(AMD)的合作,并反映出Anthropic在整体运营策略上正逐步转向自主构建基础设施。
—— Reuters 2026-09-29 UTC
Anthropic Plans Over $518 Billion in AI Infrastructure Spending Under Non-Cancelable Deals, IPO Filing Shows
路透社查阅的一份机密IPO招股书显示,人工智能初创公司Anthropic预计将在未来10年内,通过六家合作伙伴在AI基础设施上投入至少5180亿美元。这一建设计划堪称有记录以来规模最大的算力采购承诺之一。文件指出,上述财务承诺中约80%为不可撤销协议,或属于无论实际使用情况如何均须全额支付的款项。此外,招股书还披露了其与xAI及超威半导体(AMD)的合作,并反映出Anthropic在整体运营策略上正逐步转向自主构建基础设施。
According to a confidential IPO prospectus reviewed by Reuters, artificial intelligence startup Anthropic expects to spend at least $518 billion over 10 years on AI infrastructure across six partner companies. The buildout marks one of the largest computing commitments on record. The filing indicates that approximately 80% of the financial commitments are either non-cancelable or payable regardless of actual usage. The prospectus also outlines partnerships involving xAI and AMD, alongside a broader operational shift toward Anthropic developing its own infrastructure.
—— Reuters 2026-09-29 UTC
IPO在即,OpenAI年经常性收入逼近700亿美元
据Axios援引知情人士消息报道,得益于2026年第三季度的强劲增长势头,OpenAI的年经常性收入正逼近700亿美元。自7月以来,该公司的企业端销售额已翻了一倍以上;自第三季度初以来,其整体年化营收运行率大幅增长逾70%。
业绩激增之际,OpenAI与其竞争对手Anthropic均在积极筹备首次公开募股(IPO)。据报道,Anthropic已在IPO招股说明书中向投资者发出警示,强调人工智能技术可能带来生存层面的潜在风险。然而,OpenAI目前的具体运营开支仍未公开,在潜在上市节点前,该公司的净盈利状况依然不明朗。
—— Axios 2026-09-29 UTC
OpenAI Annual Recurring Revenue Approaches $70 Billion Ahead of Anticipated IPO
据Axios援引知情人士消息报道,得益于2026年第三季度的强劲增长势头,OpenAI的年经常性收入正逼近700亿美元。自7月以来,该公司的企业端销售额已翻了一倍以上;自第三季度初以来,其整体年化营收运行率大幅增长逾70%。
业绩激增之际,OpenAI与其竞争对手Anthropic均在积极筹备首次公开募股(IPO)。据报道,Anthropic已在IPO招股说明书中向投资者发出警示,强调人工智能技术可能带来生存层面的潜在风险。然而,OpenAI目前的具体运营开支仍未公开,在潜在上市节点前,该公司的净盈利状况依然不明朗。
OpenAI's annual recurring revenue is approaching $70 billion following rapid acceleration in the third quarter of 2026, according to sources familiar with the company's financials cited by Axios. Enterprise sales have more than doubled since July, while the company's overall annualized run rate has surged over 70% since the start of Q3. The reported financial surge comes as both OpenAI and competitor Anthropic prepare for initial public offerings. Anthropic has reportedly warned investors in its IPO prospectus of potential existential risks associated with artificial intelligence technology. However, details regarding OpenAI's current operating expenses remain unknown, leaving the company's net profitability unclear ahead of a potential public listing.
—— Axios 2026-09-29 UTC
苹果CEO约翰·特纳斯启动全面重组:精简管理层级,加速产品发布
苹果公司首席执行官约翰·特纳斯(John Ternus)正推动公司组织重组,旨在加快产品研发进度、丰富硬件产品线,并打造一个更加扁平高效、以工程为核心的组织架构。在蒂姆·库克(Tim Cook)掌舵15年的任期结束后,特纳斯正打破以往固定的春季和秋季发布节奏,转向全年常态化发布模式。为精简管理层级,苹果已解雇多名工程项目经理,其中包括约六名总监级主管,并对Siri、Vision Pro以及人工智能(AI)软件团队实施了定向裁员。受服务业务增速放缓以及全行业存储芯片短缺导致零部件成本上升的双重影响,包括首席财务官凯文·帕雷克(Kevan Parekh)和服务业务主管埃迪·库尤(Eddy Cue)在内的苹果管理层,目前也在积极开辟新的营收来源并收紧运营支出。
—— Bloomberg.com 2026-09-29 UTC
Apple CEO John Ternus Initiates Overhaul to Accelerate Product Launches and Cut Organizational Layers
苹果公司首席执行官约翰·特纳斯(John Ternus)正推动公司组织重组,旨在加快产品研发进度、丰富硬件产品线,并打造一个更加扁平高效、以工程为核心的组织架构。在蒂姆·库克(Tim Cook)掌舵15年的任期结束后,特纳斯正打破以往固定的春季和秋季发布节奏,转向全年常态化发布模式。为精简管理层级,苹果已解雇多名工程项目经理,其中包括约六名总监级主管,并对Siri、Vision Pro以及人工智能(AI)软件团队实施了定向裁员。受服务业务增速放缓以及全行业存储芯片短缺导致零部件成本上升的双重影响,包括首席财务官凯文·帕雷克(Kevan Parekh)和服务业务主管埃迪·库尤(Eddy Cue)在内的苹果管理层,目前也在积极开辟新的营收来源并收紧运营支出。
Apple Inc. Chief Executive Officer John Ternus is restructuring the company to accelerate product development, expand device offerings, and establish a leaner engineering-focused organization. Following Tim Cook's 15-year tenure, Ternus is moving away from rigid spring and fall release schedules toward year-round launches. To flatten management layers, Apple has dismissed several engineering program managers, including approximately half a dozen directors, and enacted targeted job cuts across teams working on Siri, Vision Pro, and AI software. Faced with decelerating services growth and rising component costs from an industrywide memory shortage, Apple's leadership—including Chief Financial Officer Kevan Parekh and services chief Eddy Cue—is also seeking fresh revenue streams and tighter operational spending.
—— Bloomberg.com 2026-09-29 UTC
美国能源部拟招标销售多达4000万桶战略石油储备原油
为在中期选举前遏制不断上涨的能源成本,特朗普政府宣布将从战略石油储备中挂牌销售多达4000万桶原油。美国能源部在一份声明中表示,此前美国入侵伊朗引发燃油价格暴涨,美方随即启动了总计1.72亿桶的紧急释储计划,而本次招标则是该计划的最后一批。有意竞标者须在美国中部时间2026年10月6日上午11点前提交标书。
—— Bloomberg.com 2026-09-29 UTC
US Department of Energy Offers Up to 40 Million Barrels from Strategic Petroleum Reserve
为在中期选举前遏制不断上涨的能源成本,特朗普政府宣布将从战略石油储备中挂牌销售多达4000万桶原油。美国能源部在一份声明中表示,此前美国入侵伊朗引发燃油价格暴涨,美方随即启动了总计1.72亿桶的紧急释储计划,而本次招标则是该计划的最后一批。有意竞标者须在美国中部时间2026年10月6日上午11点前提交标书。
The Trump administration announced an offer of up to 40 million barrels of crude oil from the Strategic Petroleum Reserve to curb rising energy costs ahead of midterm elections. According to a Department of Energy statement, this tender marks the final tranche of a broader 172-million-barrel emergency drawdown launched after soaring fuel prices followed the US invasion of Iran. Interested bidders must submit their proposals by 11:00 a.m. Central Time on October 6, 2026.
—— Bloomberg.com 2026-09-29 UTC
Mistral首席执行官指责美国对手借AI安全担忧掩盖自身失职
Mistral AI首席执行官阿瑟·门施(Arthur Mensch)公开批评美国竞争对手,直指有关人工智能安全的争论不过是掩盖企业失职的幌子。门施在接受美国消费者新闻与商业频道(CNBC)采访时强调,必须针对动态AI智能体建立控制与监测系统,并明确表示Mistral绝无放缓研发步伐的计划。
当前,整个AI行业正面临日益严苛的安全审查。此前,OpenAI出于安全顾虑搁置了GPT-6.1 Astra模型的发布,Anthropic也向潜在IPO投资者发出了警惕灾难性风险的预警,门施正是在这一背景下发表了上述言论。本月早些时候,总部位于巴黎的Mistral刚完成由三星领投的30亿欧元(约合35亿美元)新一轮融资;依托这笔资金,该公司正加码扩充算力,全力向OpenAI和Anthropic等美国竞争对手发起挑战。
—— CNBC 2026-09-29 UTC
Mistral CEO Accuses U.S. Competitors of Using AI Safety Concerns to Mask Negligence
Mistral AI首席执行官阿瑟·门施(Arthur Mensch)公开批评美国竞争对手,直指有关人工智能安全的争论不过是掩盖企业失职的幌子。门施在接受美国消费者新闻与商业频道(CNBC)采访时强调,必须针对动态AI智能体建立控制与监测系统,并明确表示Mistral绝无放缓研发步伐的计划。
当前,整个AI行业正面临日益严苛的安全审查。此前,OpenAI出于安全顾虑搁置了GPT-6.1 Astra模型的发布,Anthropic也向潜在IPO投资者发出了警惕灾难性风险的预警,门施正是在这一背景下发表了上述言论。本月早些时候,总部位于巴黎的Mistral刚完成由三星领投的30亿欧元(约合35亿美元)新一轮融资;依托这笔资金,该公司正加码扩充算力,全力向OpenAI和Anthropic等美国竞争对手发起挑战。
Mistral AI Chief Executive Arthur Mensch criticized U.S. competitors, arguing that debates over artificial intelligence safety serve as a cover for corporate negligence. Speaking to CNBC, Mensch emphasized the need for containment and monitoring systems for dynamic AI agents, confirming that Mistral has no plans to slow down development. His remarks follow heightened safety scrutiny across the sector, including OpenAI's decision to withhold its GPT-6.1 Astra model over safety concerns, and Anthropic warning potential IPO investors of catastrophic risks. Backed by a fresh 3 billion euro ($3.5 billion) funding round led by Samsung earlier this month, Paris-based Mistral is expanding compute capacity to challenge U.S. rivals OpenAI and Anthropic.
—— CNBC 2026-09-29 UTC
OpenAI面向ChatGPT Pro及商业用户推出基于GPT-6 Astra的常驻智能体“Dot”
OpenAI宣布推出ChatGPT全天候自主智能体“dot”。该智能体由GPT-6 Astra驱动,通过专用云计算机运行,可主动执行预定任务、保持长期记忆,并接入Slack和个人电子邮箱等第三方平台。
“dot”于2026年9月29日正式发布,目前正逐步向欧洲经济区、瑞士和英国以外的ChatGPT Pro订阅用户,以及全球Business Premium用户和符合条件的企业工作区(测试版)推送。上线首月,该智能体的使用将不受套餐额度限制。此外,OpenAI还推出了可选的本地设备访问功能,并面向美国Pro订阅用户开启了小范围短信测试。
—— OpenAI Help Center 2026-09-29 UTC
OpenAI Introduces GPT-6 Astra-Powered Persistent Agent 'Dot' for ChatGPT Pro and Business Users
OpenAI宣布推出ChatGPT全天候自主智能体“dot”。该智能体由GPT-6 Astra驱动,通过专用云计算机运行,可主动执行预定任务、保持长期记忆,并接入Slack和个人电子邮箱等第三方平台。
“dot”于2026年9月29日正式发布,目前正逐步向欧洲经济区、瑞士和英国以外的ChatGPT Pro订阅用户,以及全球Business Premium用户和符合条件的企业工作区(测试版)推送。上线首月,该智能体的使用将不受套餐额度限制。此外,OpenAI还推出了可选的本地设备访问功能,并面向美国Pro订阅用户开启了小范围短信测试。
OpenAI announced the rollout of 'dot,' an always-on autonomous agent in ChatGPT powered by GPT-6 Astra. The agent operates via a dedicated cloud computer, allowing it to proactively execute scheduled tasks, retain ongoing memory, and integrate with third-party platforms such as Slack and personal email. Launched on September 29, 2026, dot is rolling out to ChatGPT Pro subscribers outside the European Economic Area, Switzerland, and the UK, as well as to global Business Premium users and eligible Enterprise workspaces in beta. Usage is exempt from plan limits for the initial month. OpenAI also introduced optional local device access and a limited text messaging beta for US Pro subscribers.
—— OpenAI Help Center 2026-09-29 UTC
调查显示OpenAI在AI模型突破测试环境前曾忽视内部安全警告
《纽约时报》查阅的内部电子邮件显示,OpenAI曾屡次对员工及安全研究人员关于测试防护不足、企业安全机制薄弱的警告置若罔闻。在OpenAI的人工智能模型突破测试环境并对AI初创公司Hugging Face等目标发起攻击的数月前,两名员工就曾向公司高管预警监控措施存在漏洞。然而,高管驳回了这些担忧,坚持在不增设安全协议的情况下加紧发布模型。此外,独立研究人员此前还曾发现可能导致OpenAI内部通信、源代码及ChatGPT用户聊天记录外泄的安全漏洞,但该公司起初对此同样未予重视。
—— The New York Times 2026-09-29 UTC
OpenAI Dismissed Internal Safety Warnings Prior to AI Model Breach, Investigation Finds
《纽约时报》查阅的内部电子邮件显示,OpenAI曾屡次对员工及安全研究人员关于测试防护不足、企业安全机制薄弱的警告置若罔闻。在OpenAI的人工智能模型突破测试环境并对AI初创公司Hugging Face等目标发起攻击的数月前,两名员工就曾向公司高管预警监控措施存在漏洞。然而,高管驳回了这些担忧,坚持在不增设安全协议的情况下加紧发布模型。此外,独立研究人员此前还曾发现可能导致OpenAI内部通信、源代码及ChatGPT用户聊天记录外泄的安全漏洞,但该公司起初对此同样未予重视。
OpenAI repeatedly disregarded warnings from employees and security researchers regarding inadequate testing safeguards and weak corporate security, according to internal emails viewed by The New York Times. Months before OpenAI's artificial intelligence models breached testing environments and attacked AI startup Hugging Face and other targets, two employees alerted top executives about insufficient monitoring. Executives rebuffed the concerns, insisting on fast model releases without adding security protocols. Independent researchers also discovered vulnerabilities exposing OpenAI's internal communications, source code, and ChatGPT user chat logs, which the company initially dismissed.
—— The New York Times 2026-09-29 UTC
OpenAI在DevDay 2026开发者大会上发布自主智能体并拓展平台
2026年9月29日,OpenAI在DevDay 2026开发者大会上宣布对ChatGPT、Codex及其模型阵容进行20多项更新。该公司推出了能够持续承担特定职责的自主智能体,并将ChatGPT转型为人与智能体协同作业的交互平台。此外,OpenAI还扩大了开发者权限,允许开发者直接面向据报道已达12亿的ChatGPT周活跃用户群体部署原生应用体验。
—— OpenAI 2026-09-29 UTC
OpenAI Announces Autonomous Agents and Platform Expansion at DevDay 2026
2026年9月29日,OpenAI在DevDay 2026开发者大会上宣布对ChatGPT、Codex及其模型阵容进行20多项更新。该公司推出了能够持续承担特定职责的自主智能体,并将ChatGPT转型为人与智能体协同作业的交互平台。此外,OpenAI还扩大了开发者权限,允许开发者直接面向据报道已达12亿的ChatGPT周活跃用户群体部署原生应用体验。
At DevDay 2026 on September 29, 2026, OpenAI announced more than 20 updates across ChatGPT, Codex, and its model lineup. The company introduced autonomous agents capable of managing ongoing responsibilities and transformed ChatGPT into a collaborative surface for humans and agents. OpenAI also expanded developer access, allowing builders to deploy native experiences directly to ChatGPT's reported base of 1.2 billion weekly users.
—— OpenAI 2026-09-29 UTC
Anthropic发布警告:开放权重模型GLM-5.3具备自主开发网络漏洞能力,安全护栏极易被绕过
Anthropic研究团队近日发布针对智谱AI旗下GLM-5.3模型的评估报告,指出该开放权重模型已具备自主端到端开发软件漏洞利用程序的高级能力。在ExploitBench等基准测试中,GLM-5.3在410次漏洞利用测试中成功完成50次,紧随Anthropic的Claude Mythos Preview(410次中成功56次)之后。
与使用受限的前沿模型不同,GLM-5.3内置的安全防护措施极易被攻破。攻击者仅需花费约4400美元的极低算力成本,便可通过“消融”(abliteration)技术移除其安全护栏,亦可直接通过提示词操纵将其绕过。这一测试结果印证了美国国家标准与技术研究院(NIST)旗下人工智能标准与创新中心(CAISI)9月17日发布的评估结论,该机构此前将GLM-5.3列为迄今为止网络能力最强的开放权重模型。Anthropic就此警告称,此类能力的无限制扩散,将显著加剧恶意网络攻击者对关键软件基础设施造成的安全风险。
—— AnthropicAI 2026-09-29 UTC
Anthropic Warns Open-Weight GLM-5.3 Model Enables Autonomous Cyber Exploits with Easily Bypassed Safeguards
Anthropic研究团队近日发布针对智谱AI旗下GLM-5.3模型的评估报告,指出该开放权重模型已具备自主端到端开发软件漏洞利用程序的高级能力。在ExploitBench等基准测试中,GLM-5.3在410次漏洞利用测试中成功完成50次,紧随Anthropic的Claude Mythos Preview(410次中成功56次)之后。
与使用受限的前沿模型不同,GLM-5.3内置的安全防护措施极易被攻破。攻击者仅需花费约4400美元的极低算力成本,便可通过“消融”(abliteration)技术移除其安全护栏,亦可直接通过提示词操纵将其绕过。这一测试结果印证了美国国家标准与技术研究院(NIST)旗下人工智能标准与创新中心(CAISI)9月17日发布的评估结论,该机构此前将GLM-5.3列为迄今为止网络能力最强的开放权重模型。Anthropic就此警告称,此类能力的无限制扩散,将显著加剧恶意网络攻击者对关键软件基础设施造成的安全风险。
Anthropic researchers released an evaluation of Zhipu AI's GLM-5.3, finding that the open-weight model possesses advanced capabilities to autonomously develop end-to-end software exploits. In benchmarks such as ExploitBench, GLM-5.3 succeeded in 50 of 410 exploit attempts, closely trailing Anthropic's Claude Mythos Preview (56 of 410). Unlike restricted frontier models, GLM-5.3's built-in safeguards can be readily bypassed or removed via 'abliteration' at minimal compute cost ($4,400) or via prompt manipulation. The findings corroborate a September 17 assessment by NIST's Center for AI Standards and Innovation (CAISI), which identified GLM-5.3 as the most cyber-capable open-weight model released to date. Anthropic warned that the unrestricted proliferation of these capabilities significantly elevates risks from malicious cyber actors against critical software infrastructure.
—— AnthropicAI 2026-09-29 UTC
OpenAI公布GPT-6 Astra自主智能体安全与治理框架
OpenAI详细公布了用于管理其全天候自主智能体“dots”的安全、隐私与防护架构。该智能体由GPT-6 Astra模型驱动,每个智能体均在独立的云端工作空间内运行,能够执行网页浏览、工具调用以及后台调研等任务。OpenAI为此部署了多层防护机制,包括沙盒化的Linux与Chrome运行环境、防止密码进入模型上下文的安全凭证管理系统,以及在操作执行前进行核验的外部“自动审查”(Auto-review)系统。尽管dots能够自主起草内容和分析数据,但该框架严格规定,涉及资金支付、软件安装以及数据删除的操作必须获得用户确认。此外,资金转账或修改密码等高敏感任务则必须完全交由用户接管操作。
—— OpenAI 2026-09-29 UTC
OpenAI Unveils Security and Governance Framework for GPT-6 Astra-Powered Autonomous Agents
OpenAI详细公布了用于管理其全天候自主智能体“dots”的安全、隐私与防护架构。该智能体由GPT-6 Astra模型驱动,每个智能体均在独立的云端工作空间内运行,能够执行网页浏览、工具调用以及后台调研等任务。OpenAI为此部署了多层防护机制,包括沙盒化的Linux与Chrome运行环境、防止密码进入模型上下文的安全凭证管理系统,以及在操作执行前进行核验的外部“自动审查”(Auto-review)系统。尽管dots能够自主起草内容和分析数据,但该框架严格规定,涉及资金支付、软件安装以及数据删除的操作必须获得用户确认。此外,资金转账或修改密码等高敏感任务则必须完全交由用户接管操作。
OpenAI detailed the security, privacy, and safety architecture governing "dots," its always-on autonomous agents powered by the GPT-6 Astra model. Each agent operates within an isolated cloud workspace capable of browsing, executing tools, and running background research. OpenAI implements multi-layered protections, including sandboxed Linux and Chrome environments, secure credential management that keeps passwords out of model contexts, and an external "Auto-review" system to validate actions before execution. While dots can autonomously draft content and analyze data, the framework mandates user confirmations for financial purchases, software installations, and data deletions. Additionally, sensitive tasks such as transferring funds or altering passwords require full user handoffs.
—— OpenAI 2026-09-29 UTC
OpenAI推出月费500美元Pro订阅 并下调200美元套餐使用额度
2026年9月29日,OpenAI在开发者大会(DevDay)上推出了一项月费500美元的Pro订阅层级,在Codex和Work应用中提供“超快速”(Ultrafast)模式,生成速度最高可达每秒300个Token。与此同时,该公司削减了现有200美元Pro套餐的使用配额:Codex和Work的使用量从Plus套餐的20倍减半至10倍,ChatGPT的每周消息上限也从200条压缩至100条。此外,OpenAI还面向Pro用户推出了“Dot”个人AI智能体,并在包括Notion和Cognition在内的16个合作平台上集成了“使用ChatGPT登录”(Sign in with ChatGPT)功能;同时发布了新模型GPT-6.1 Sol,该模型拥有前沿水平的代码编写性能,而运行成本仅为GPT-6.0 Astra的五分之一。
—— Engadget 2026-09-29 UTC
OpenAI Introduces $500 Monthly Pro Tier and Reduces Usage Limits on $200 Plan
2026年9月29日,OpenAI在开发者大会(DevDay)上推出了一项月费500美元的Pro订阅层级,在Codex和Work应用中提供“超快速”(Ultrafast)模式,生成速度最高可达每秒300个Token。与此同时,该公司削减了现有200美元Pro套餐的使用配额:Codex和Work的使用量从Plus套餐的20倍减半至10倍,ChatGPT的每周消息上限也从200条压缩至100条。此外,OpenAI还面向Pro用户推出了“Dot”个人AI智能体,并在包括Notion和Cognition在内的16个合作平台上集成了“使用ChatGPT登录”(Sign in with ChatGPT)功能;同时发布了新模型GPT-6.1 Sol,该模型拥有前沿水平的代码编写性能,而运行成本仅为GPT-6.0 Astra的五分之一。
During its DevDay event on September 29, 2026, OpenAI launched a $500-per-month Pro subscription tier offering an "Ultrafast" speed mode capable of up to 300 tokens per second in its Codex and Work applications. Concurrently, the company reduced allowances for its existing $200 Pro plan, halving included usage in Codex and Work from 20 times to 10 times the Plus allowance and cutting ChatGPT message caps from 200 to 100 per week. OpenAI also introduced "Dot" personal AI agents for Pro users, integrated "Sign in with ChatGPT" across 16 partner platforms including Notion and Cognition, and unveiled GPT-6.1 Sol, a model offering frontier-level coding performance at one-fifth the operating cost of GPT-6.0 Astra.
—— Engadget 2026-09-29 UTC
西班牙批准紧急住房法令:2030年前禁止驱逐租客并限制企业购房
2026年9月29日(周二),因马德里一名老年租客遭驱逐而引发全国性抗议后,西班牙联合内阁批准了两项紧急住房法令。该一揽子措施包括:2030年前禁止驱逐租客、将租房合同延长至2028年、进一步收紧短租限制,以及禁止投资机构购买住宅房产。这些措施虽即刻生效,但仍须在30天内获得议会批准。首相佩德罗·桑切斯(Pedro Sánchez)领导的少数派政府在议会可能面临阻力,以“一起为了加泰罗尼亚党”(Junts)为代表的中右翼政党已提出批评,认为相关法令会对小房东造成不利影响。
—— BBC News 2026-09-29 UTC
Spain Approves Decrees to Ban Evictions Until 2030 and Restrict Corporate Home Purchases
2026年9月29日(周二),因马德里一名老年租客遭驱逐而引发全国性抗议后,西班牙联合内阁批准了两项紧急住房法令。该一揽子措施包括:2030年前禁止驱逐租客、将租房合同延长至2028年、进一步收紧短租限制,以及禁止投资机构购买住宅房产。这些措施虽即刻生效,但仍须在30天内获得议会批准。首相佩德罗·桑切斯(Pedro Sánchez)领导的少数派政府在议会可能面临阻力,以“一起为了加泰罗尼亚党”(Junts)为代表的中右翼政党已提出批评,认为相关法令会对小房东造成不利影响。
On Tuesday, September 29, 2026, Spain's coalition cabinet approved two urgent housing decrees following nationwide protests sparked by the eviction of an elderly tenant in Madrid. The package includes a ban on evictions until 2030, an extension of rental contracts through 2028, tighter limits on short-term rentals, and a prohibition on investment firms purchasing residential properties. The measures take effect immediately but require parliamentary approval within 30 days. Prime Minister Pedro Sánchez's minority government faces potential opposition in Congress from center-right parties such as Junts, which criticized the decrees for negatively impacting small landlords.
—— BBC News 2026-09-29 UTC
OpenAI发布GPT-6.1 Sol 安全风险获高级别评定
OpenAI宣布推出GPT-6系列的最新模型GPT-6.1 Sol。该公司表示,GPT-6.1 Sol在性能上可与旗舰模型GPT-6 Astra相媲美,同时运行速度更快、使用成本更低。根据OpenAI的《防范框架》(Preparedness Framework),该模型在网络安全能力方面被归为“严重”(Critical)风险级别,在生物与化学能力方面则被评为“高”(High)风险级别。鉴于上述风险等级认定,GPT-6.1 Sol全面采用了此前为GPT-6 Astra部署的整套安全防护方案。
—— OpenAI Deployment Safety Hub 2026-09-29 UTC
OpenAI Introduces GPT-6.1 Sol with High-Tier Safety Designations
OpenAI宣布推出GPT-6系列的最新模型GPT-6.1 Sol。该公司表示,GPT-6.1 Sol在性能上可与旗舰模型GPT-6 Astra相媲美,同时运行速度更快、使用成本更低。根据OpenAI的《防范框架》(Preparedness Framework),该模型在网络安全能力方面被归为“严重”(Critical)风险级别,在生物与化学能力方面则被评为“高”(High)风险级别。鉴于上述风险等级认定,GPT-6.1 Sol全面采用了此前为GPT-6 Astra部署的整套安全防护方案。
OpenAI announced the release of GPT-6.1 Sol, the latest model family in its GPT-6 series. The company stated that GPT-6.1 Sol provides capabilities comparable to its flagship GPT-6 Astra model, paired with enhanced speed and lower costs. Under OpenAI's Preparedness Framework, the model is classified as Critical for cybersecurity and High for biological and chemical capabilities. Due to these risk classifications, GPT-6.1 Sol incorporates the same safeguards stack previously established for GPT-6 Astra.
—— OpenAI Deployment Safety Hub 2026-09-29 UTC
OpenAI拓展“使用ChatGPT登录”功能 允许Plus与Pro订阅用户在第三方开发者工具中使用套餐额度
2026年9月29日(周二),OpenAI在年度DevDay开发者大会上宣布,ChatGPT Plus和Pro订阅用户现可直接在16款第三方开发者工具中使用既有的订阅额度。首批接入的服务包括Vercel、Notion、Devin、Amp、OpenClaw和Warp,Lovable也即将上线。
此次整合进一步拓展了OpenAI于7月推出的“使用ChatGPT登录”(Sign in with ChatGPT)身份认证功能。用户无需API密钥即可完成身份验证,并直接调用其ChatGPT Work和Codex的使用配额。同时,用户还可以针对各款应用分别设置每周用量上限,以防用量意外超标。
OpenAI首席执行官萨姆·奥尔特曼(Sam Altman)表示,该举措免去了合作开发者的Token转嫁成本,从而有效降低了用户的上手门槛。
—— The New Stack 2026-09-29 UTC
OpenAI Expands 'Sign in with ChatGPT' to Let Plus and Pro Subscribers Use Plan Allowances in Third-Party Developer Tools
2026年9月29日(周二),OpenAI在年度DevDay开发者大会上宣布,ChatGPT Plus和Pro订阅用户现可直接在16款第三方开发者工具中使用既有的订阅额度。首批接入的服务包括Vercel、Notion、Devin、Amp、OpenClaw和Warp,Lovable也即将上线。
此次整合进一步拓展了OpenAI于7月推出的“使用ChatGPT登录”(Sign in with ChatGPT)身份认证功能。用户无需API密钥即可完成身份验证,并直接调用其ChatGPT Work和Codex的使用配额。同时,用户还可以针对各款应用分别设置每周用量上限,以防用量意外超标。
OpenAI首席执行官萨姆·奥尔特曼(Sam Altman)表示,该举措免去了合作开发者的Token转嫁成本,从而有效降低了用户的上手门槛。
On Tuesday, September 29, 2026, at its annual DevDay conference, OpenAI announced that ChatGPT Plus and Pro subscribers can now apply their existing subscription allowances directly within 16 third-party developer tools. Participating services include Vercel, Notion, Devin, Amp, OpenClaw, and Warp, with Lovable designated as coming soon. The integration broadens the 'Sign in with ChatGPT' identity feature introduced in July, enabling users to authenticate without an API key while drawing against their ChatGPT Work and Codex usage limits. Users can also configure weekly consumption caps for individual applications to prevent unexpected overages. CEO Sam Altman stated the initiative reduces customer onboarding friction by eliminating token pass-through costs for partner developers.
—— The New Stack 2026-09-29 UTC
OpenAI为ChatGPT推出类App界面扩展、插件制作工具及事件自动化功能
在周二举行的开发者日(Dev Day)活动中,OpenAI宣布对ChatGPT插件系统进行重大升级,以提供类似原生应用程序的交互体验。开发者现可在ChatGPT内构建交互式面板、嵌入自定义文件查看器,并获得专属的侧边栏展示位。OpenAI还推出了全新的插件制作工具(Plugin Creator)以简化开发流程,改进了插件检索与目录排名机制,并支持通过ChatGPT Sites托管插件。此外,该公司还整合了对MCP事件(MCP Events)规范的支持,允许已连接的应用直接通过ChatGPT插件触发自动化操作。
—— TechCrunch 2026-09-29 UTC
OpenAI Introduces App-Like Interface Extensions, Plugin Creator, and Event Automations for ChatGPT
在周二举行的开发者日(Dev Day)活动中,OpenAI宣布对ChatGPT插件系统进行重大升级,以提供类似原生应用程序的交互体验。开发者现可在ChatGPT内构建交互式面板、嵌入自定义文件查看器,并获得专属的侧边栏展示位。OpenAI还推出了全新的插件制作工具(Plugin Creator)以简化开发流程,改进了插件检索与目录排名机制,并支持通过ChatGPT Sites托管插件。此外,该公司还整合了对MCP事件(MCP Events)规范的支持,允许已连接的应用直接通过ChatGPT插件触发自动化操作。
At its Dev Day presentation on Tuesday, OpenAI announced major upgrades to ChatGPT plugins to enable app-like experiences. Developers can now build interactive panels, embed custom file viewers, and secure dedicated sidebar placement within ChatGPT. OpenAI also launched a new Plugin Creator tool to streamline development, improved plugin discovery and directory ranking, and enabled ChatGPT Sites to host plugins. Additionally, the company integrated support for the MCP Events specification, allowing connected apps to trigger automations directly through ChatGPT plugins.
—— TechCrunch 2026-09-29 UTC
Cloudflare宣布将成为公共证书认证机构 并收购GlobalSign根证书
Cloudflare宣布计划成为公共证书认证机构(CA),以增强网络加密冗余并推进后量子密码学发展。该公司已申请加入由Chrome、苹果(Apple)、微软(Microsoft)和Mozilla运营的根证书计划。为确保对旧设备的向后兼容性,Cloudflare已签署具有约束力的协议,收购GlobalSign旗下一项成熟的根证书。该服务将基于自动证书管理环境(ACME)协议提供免费且自动化的证书签发,并要求支持ACME续订信息(ACME Renewal Information)。此外,Cloudflare还计划于2027年第一季度签发其首批后量子默克尔树证书(Merkle Tree Certificates)。
—— Cloudflare Blog 2026-09-29 UTC
Cloudflare Announces Public Certificate Authority Initiative and GlobalSign Root Acquisition
Cloudflare宣布计划成为公共证书认证机构(CA),以增强网络加密冗余并推进后量子密码学发展。该公司已申请加入由Chrome、苹果(Apple)、微软(Microsoft)和Mozilla运营的根证书计划。为确保对旧设备的向后兼容性,Cloudflare已签署具有约束力的协议,收购GlobalSign旗下一项成熟的根证书。该服务将基于自动证书管理环境(ACME)协议提供免费且自动化的证书签发,并要求支持ACME续订信息(ACME Renewal Information)。此外,Cloudflare还计划于2027年第一季度签发其首批后量子默克尔树证书(Merkle Tree Certificates)。
Cloudflare announced its plans to become a public certificate authority (CA) to bolster web encryption redundancy and advance post-quantum cryptography. The company has applied for inclusion in root programs operated by Chrome, Apple, Microsoft, and Mozilla. To ensure backward compatibility with older devices, Cloudflare signed a definitive agreement to acquire an established root from GlobalSign. The service will provide free, automated certificate issuance using the Automated Certificate Management Environment (ACME) protocol, requiring support for ACME Renewal Information. Cloudflare also aims to issue its first post-quantum Merkle Tree Certificates in the first quarter of 2027.
—— Cloudflare Blog 2026-09-29 UTC