🏆 Headline
G7 AI Four-Heavy Lunch: Call for US-Dominated AI Governance Alliance
# AI Daily News | 2026年6月18日 On June 17, the final day of the G7 Summit in Evian, France, global AI lab CEOs — Sam Altman of OpenAI, Dario Amodei of Anthropic, Demis Hassabis of Google DeepMind, and Arthur Mensch of Mistral AI — joined G7 leaders for a two-and-a-half-hour closed-door lunch. U.S. President Trump, French President Macron, German Chancellor Merz, and others were all in attendance, marking AI's formal elevation from a tech topic to a core item of global geopolitics. During the lunch, Amodei and Hassabis independently proposed the concept of establishing a "U.S.-led AI alliance." Amodei suggested building an international cooperation framework for tiered access to frontier AI models and critical chip trade. Hassabis added that countries should deepen coordination in high-risk areas such as cybersecurity and biosecurity. Altman's remarks were relatively mild; he called for an "International AI Testing Forum" to provide a platform for globally accepted capability assessments and risk analysis, acting as a bridge for cooperation between nations. The background to this lunch meeting was unusually sensitive: just five days prior, the U.S. government, citing national security, had banned non-U.S. citizens from accessing Anthropic's latest models, Fable 5 and Mythos 5. This move triggered strong backlash in the EU, with leaders in France and the EU publicly calling for reduced reliance on U.S. AI. At the lunch, Trump sat next to Altman, while Amodei sat across — the seating arrangement itself was interpreted as a political signal. Macron told reporters after the meeting: "The challenge we face is frontier models." Synthesia CEO Victor Riparbelli was even more blunt after the event, telling POLITICO: "Everyone knows G7 nations and allies must unite to win the AI race; we cannot let non-democratic nations gain the upper hand." This suggests the competitive logic behind AI governance frameworks — that establishing standards is itself a geopolitical game.
Source: CNBC / Politico / Fortune / Reuters | 2026-06-17
Anthropic becomes first AI company to join Frontier Carbon Removal Alliance
Frontier, the carbon removal procurement alliance co-founded by Stripe and Google in 2022, announced a new round of $915 million funding, bringing total committed capital to nearly $1.8 billion. More importantly, Anthropic became the first all-AI company member of the alliance. Frontier plans to focus on 10 to 15 most promising carbon removal tech companies through 8 to 10-year long-term purchase contracts. Anthropic stated that the U.S. AI industry may need at least 50GW of new power capacity by 2028. This is Anthropic's first public climate action plan; the company had previously kept a low profile on corporate sustainability.
Source: TechCrunch / Reuters / Axios | 2026-06-17
World Model Company Odyssey Raises $310M Series B at $1.45B Valuation
AI lab Odyssey, founded by autonomous driving pioneers Oliver Cameron and Jeff Hawke, announced completion of a $310 million Series B round, valuing the company at $1.45 billion. The round was led by Natural Capital, with participation from Amazon, AMD Ventures, Google Ventures, EQT, and In-Q-Tel (CIA-affiliated). Odyssey focuses on building a general "world model" — a foundational model capable of simulating the 3D workings of the physical world, considered the next frontier after large language models. The company also partnered with AWS to gain access to its self-developed Trainium chips. Competition in the world model track is currently fierce: Yann LeCun's AMI Labs launched a funding round at a €3 billion valuation, Runway reached $40 million quarterly ARR after launching products in December, and Google DeepMind's Genie model is already applied at Waymo.
Source: TechCrunch / SiliconANGLE / BusinessWire | 2026-06-17
Bezos' VivaTech Debut: AI Won't Cause Mass Unemployment, It Will Cause Labor Shortages
Amazon founder and world's fourth-richest man Jeff Bezos, speaking at the opening ceremony of the 10th-anniversary VivaTech conference in Paris, proposed a counter-intuitive point: AI will not cause mass unemployment; instead, it will cause labor shortages. He argued that human desires are "infinite," and AI's lowered barriers to implementation will unleash more creativity and demand, rather than replacing humans. Bezos also introduced Prometheus, an AI startup he co-founded at the end of 2025, which focuses on using AI to accelerate physical manufacturing and industrial system design. The company reportedly raised $12 billion with a valuation of $41 billion — with no commercial products launched yet. Bezos also mentioned Blue Origin's space vision, advocating for moving heavy-pollution industries to space to protect Earth.
Source: Reuters / BNN Bloomberg / VivaTech Official | 2026-06-17
Nvidia Research: AI Coding Agents Can Self-Train Robots to Install GPUs and Cut Zip Ties
Nvidia's research team demonstrated a breakthrough: allowing teams of AI coding agents (multiple AI agents collaborating) to train robots to perform complex manipulation tasks, including installing GPUs on motherboards and cutting zip ties. This marks that the capability boundaries of AI coding agents are extending from the pure software realm into the physical world. Research shows that AI coding agents can write robot training code, adjust parameters, iterate, and optimize with almost no human intervention. This offers a completely new approach to solving the "long-tail operation" problem in robot training — using AI to train physical machines controlled by AI.
Source: Ars Technica / Nvidia Research | 2026-06-17
Snap Releases $2,195 AR Glasses, Stock Plummets
Snap released its new generation AR glasses, Specs, at VivaTech in Paris, with a staggering price of $2,195. CEO Evan Spiegel declared on stage that "glasses are the next generation of computer," but the market reaction was cold — Snap's stock immediately plummeted upon the announcement. The glasses feature augmented reality display and AI functions, but the price far exceeds the affordability of the average consumer. Compared to Meta's Ray-Ban smart glasses (around $300) and Apple Vision Pro ($3,499 but a completely different functionality), Snap attempted to find a differentiated positioning in the consumer AR space, but the $2,195 price tag led most analysts to question its market viability.
Source: TechCrunch / Bloomberg | 2026-06-17
"Dangerous" AI Models Have Become an Irresistible Trend
WIRED published an in-depth report stating that AI models with advanced hacking capabilities are becoming industry normal. As major labs compete to enhance their models' reasoning and code generation capabilities, AI systems capable of autonomously discovering and exploiting software vulnerabilities have shifted from theoretical risks to reality. The report cited warnings from multiple security researchers that current AI safety assessment frameworks lag far behind the actual growth speed of model capabilities. This issue has become more urgent following Anthropic's restriction on exporting the Mythos model — reportedly possessing breakthrough capabilities in cybersecurity but also triggering intense debate on the balance between AI capability and security.
Source: WIRED / Ars Technica | 2026-06-17
Roelof Botha Joins SpaceX Board, Signal Is Clear
Legendary Sequoia Capital partner and former head of YouTube, Instagram, and Square investments, Roelof Botha, officially joined the SpaceX board. This move came just one week after SpaceX's IPO and was the company's first major governance restructuring following its acquisition of Cursor. Botha's addition is widely interpreted as a signal of SpaceX's strategic pivot to an "AI-first" approach — he possesses deep investment experience and a network of connections in consumer tech and enterprise software. Meanwhile, Thrive Capital's combined holdings in SpaceX and Cursor have exceeded $10 billion.
Source: TechCrunch | 2026-06-17
Pew Survey: Only 16% of Americans Think AI Has a Positive Impact on Society
The Pew Research Center released a latest survey report showing that only 16% of Americans believe AI will have a positive impact on society, a figure that has continued to decline over the past year. At the same time, about two-thirds of respondents believe the development speed of AI is too fast. Despite the rising usage rates of AI tools like ChatGPT, public mistrust is growing simultaneously. The survey also found that while AI users have better experiences in scenarios like education and writing assistance, they generally hold a pessimistic view on AI's impact on employment, privacy, and information security.
Source: TechCrunch / Pew Research | 2026-06-17
Poland Invests $11 Million in ElevenLabs to Build AI Tech Hub
Through its National Development Bank BGK's investment arm, Vinci SA, Poland invested $11 million in AI voice company ElevenLabs, aiming to position Poland as a major European AI technology hub. This move reflects the increasingly obvious trend of "state capitalism" in the AI field — governments are no longer just formulating AI policies but are directly investing in AI companies through sovereign funds. Previously, France committed €655 million for sovereign AI, Indian AI company Sarvam secured $234 million in funding to become a unicorn, and the Netherlands launched its sovereign language model GPT-NL. Europe is putting real money into practicing "AI sovereignty."
Source: Bloomberg | 2026-06-17
🏆 今日头条
G7峰会AI四巨头午餐会：呼吁建立美国主导的AI治理联盟
6月17日，在法国埃维昂举行的G7峰会最后一天，全球四大AI实验室的CEO——OpenAI的Sam Altman、Anthropic的Dario Amodei、Google DeepMind的Demis Hassabis、以及Mistral AI的Arthur Mensch——与G7国家领导人共进了长达两个半小时的闭门午餐会。美国总统特朗普、法国总统马克龙、德国总理默茨等悉数出席，这标志着AI已经从科技议题正式上升为全球地缘政治的核心议程。 午餐会上，Amodei和Hassabis不约而同地提出了建立"美国主导的AI联盟"的构想。Amodei建议在前沿AI模型的分级访问和关键芯片贸易领域建立国际合作框架。Hassabis则补充指出，各国应在网络安全、生物安全等高风险领域展开深度协调。Altman的发言相对温和，他呼吁建立一个"国际AI测试论坛"，为全球范围内接受的能力评估和风险分析提供平台，并充当各国合作的桥梁。 这场午餐会的背景异常敏感：就在五天前，美国政府刚刚以国家安全为由，禁止非美国公民访问Anthropic的最新模型Fable 5和Mythos 5。这一举措在欧盟引发了强烈反弹，法国和欧盟多名领导人公开呼吁减少对美国AI的依赖。在午餐会上，特朗普坐在Altman旁边，而Amodei则坐在对面——座位安排本身就被解读为一种政治信号。 马克龙在会后对记者表示："我们面临的挑战是前沿模型。"多名参会CEO在会后强调了G7国家在AI竞赛中团结协作的重要性。这番讨论暗示了AI治理框架背后的竞争逻辑——制定标准本身就是一场全球博弈。 > 💬 四个CEO加上七个国家元首，一顿午餐就决定了AI世界的"游戏规则"。民主的参与权呢？
来源：CNBC / Politico / Fortune / Reuters | 2026-06-17
Anthropic成为首家加入碳移除联盟Frontier的AI公司
由Stripe和Google于2022年联合创立的碳移除采购联盟Frontier宣布完成新一轮9.15亿美元注资，总承诺资金接近18亿美元。更重要的是，Anthropic成为该联盟中首家纯AI公司成员。Frontier计划通过8到10年的长期采购合同，集中押注10到15家最具潜力的碳移除技术公司。Anthropic表示，美国AI行业到2028年可能需要至少50GW的新增电力容量。这是Anthropic首个公开的气候行动计划，此前该公司在企业可持续发展方面一直保持低调。 > 💬 每训练一个模型要烧掉一座小城的电量，现在终于想起来买碳汇了——亡羊补牢，聊胜于无。
来源：TechCrunch / Reuters / Axios | 2026-06-17
世界模型公司Odyssey获3.1亿美元B轮融资，估值14.5亿美元
由自动驾驶先驱Oliver Cameron和Jeff Hawke创立的AI实验室Odyssey宣布完成3.1亿美元B轮融资，估值达到14.5亿美元。本轮融资由Natural Capital领投，Amazon、AMD Ventures、Google Ventures、EQT及CIA关联基金In-Q-Tel等参投。Odyssey专注于构建通用世界模型——一种能够模拟物理世界三维运作的基础模型，被视为继大语言模型之后的下一个前沿方向。公司同时与AWS达成合作，获得其自研Trainium芯片的使用权。当前世界模型赛道竞争激烈：Yann LeCun的AMI Labs以30亿欧元估值开启融资，Runway在12月推出产品后季度ARR达4000万美元，Google DeepMind的Genie模型已应用于Waymo。 > 💬 自动驾驶团队做世界模型，这赛道目前比LLM更性感——问题是"物理世界的Token"是什么？
来源：TechCrunch / SiliconANGLE / BusinessWire | 2026-06-17
贝索斯VivaTech首秀：AI不会导致大规模失业，而是劳动力短缺
亚马逊创始人、世界第四富豪Jeff Bezos在巴黎VivaTech大会十周年开幕式上发表演讲，提出了一个反直觉的观点：AI不会导致大规模失业，反而会造成劳动力短缺。他认为人类的欲望是"无穷的"，AI降低的实现门槛将释放更多创意和需求，而非取代人类。Bezos同时介绍了其2025年底联合创立的AI创业公司Prometheus，该公司专注于用AI加速物理制造和工业系统设计，据报道已融资120亿美元，估值达410亿美元——在尚未推出任何商业产品的情况下。Bezos还提及Blue Origin的太空愿景，主张将重污染产业转移到太空以保护地球。 > 💬 估值410亿、零产品、刚融了120亿——硅谷的"物理AI"叙事正在复制大模型时代的疯狂。
来源：Reuters / BNN Bloomberg / VivaTech官方 | 2026-06-17
Nvidia研究：AI编程代理可自主训练机器人安装GPU和剪扎带
Nvidia的研究团队展示了一项突破：让AI编程代理团队（即多个AI coding agent协作）来训练机器人完成复杂操作任务，包括在主板上安装GPU和剪断扎带。这标志着AI编程代理的能力边界正在从纯软件领域向物理世界延伸。研究显示，AI编程代理可以编写机器人训练代码、调整参数、迭代优化，整个过程几乎无需人类干预。这为解决机器人训练中"长尾操作"难题提供了一种全新思路——用AI来训练AI控制的物理机器。 > 💬 AI写代码训练机器人安装GPU——AI产业正在构建自己的基础设施，字面意义上的。
来源：Ars Technica / Nvidia Research | 2026-06-17
Snap发布2195美元AR眼镜，股价应声暴跌
Snap在巴黎VivaTech上发布了新一代AR眼镜Specs，售价高达2195美元。CEO Evan Spiegel在台上宣称"眼镜是下一代计算机"，但市场反应冷酷——Snap股价在消息公布后立即跳水。这款眼镜配备增强现实显示和AI功能，但价格远超普通消费者的承受范围。与Meta的Ray-Ban智能眼镜（约300美元）和Apple Vision Pro（3499美元但功能完全不同）相比，Snap试图在消费级AR领域找到差异化定位，但2195美元的定价让大多数分析师质疑其市场可行性。 > 💬 "眼镜是下一代计算机"——如果下一代计算机卖2195美元，那这代先破产了。
来源：TechCrunch / Bloomberg | 2026-06-17
"危险"AI模型已成不可阻挡趋势
WIRED发布深度报道指出，具备高级黑客能力的AI模型正在成为行业常态。随着各大实验室竞相提升模型的推理和代码生成能力，能够自主发现和利用软件漏洞的AI系统已经从理论风险变为现实。报道引述多位安全研究员的警告称，当前的AI安全评估框架远远落后于模型能力的实际增长速度。这一问题在Anthropic的Mythos模型被限制出口后变得更加紧迫——该模型据称在网络安全领域具有突破性能力，但也因此引发了关于AI能力与安全之间平衡的激烈辩论。 > 💬 每个AI公司都在比谁的模型更"危险"，然后把安全报告埋在官网第47页。
来源：WIRED / Ars Technica | 2026-06-17
Roelof Botha加入SpaceX董事会，信号明确
红杉资本传奇合伙人、曾主导投资YouTube、Instagram、Square的Roelof Botha正式加入SpaceX董事会。此举发生在SpaceX IPO后仅一周，也是该公司收购Cursor后的首次重大治理结构调整。Botha的加入被广泛解读为SpaceX向"AI优先"战略转型的信号——他在消费科技和企业软件领域拥有深厚的投资经验和人脉网络。与此同时，Thrive Capital在SpaceX和Cursor的合计持股价值已超过100亿美元。 > 💬 当一个投了半个硅谷的人加入你的董事会，说明你要么要收购更多公司，要么要被收购。
来源：TechCrunch | 2026-06-17
Pew调查：仅16%美国人认为AI对社会有积极影响
Pew Research Center发布最新调查报告，仅有16%的美国人认为AI将对社会产生积极影响，这一比例在过去一年中持续下降。与此同时，约三分之二的受访者认为AI发展速度过快。尽管ChatGPT等AI工具的使用率在上升，但公众的不信任感也在同步增长。调查还发现，AI使用者在教育、写作辅助等场景中的体验较好，但对AI在就业、隐私和信息安全方面的影响普遍持悲观态度。 > 💬 使用率在涨、信任度在跌——AI行业正在经历一场"越用越怕"的悖论。
来源：TechCrunch / Pew Research | 2026-06-17
波兰斥资1100万美元入股ElevenLabs，打造AI技术中心
波兰通过国家发展银行BGK的投资机构Vinci SA，向AI语音公司ElevenLabs投资1100万美元，旨在将波兰打造为重要的欧洲AI技术中心。这一举措反映了AI领域日益明显的"国家资本主义"趋势——各国政府不再只是制定AI政策，而是直接通过主权基金参股AI公司。此前，法国已承诺6.55亿欧元用于主权AI，印度AI公司Sarvam获2.34亿美元融资成为独角兽，荷兰推出主权语言模型GPT-NL。欧洲正在用真金白银践行"AI主权"。 > 💬 国家基金投AI语音公司——听起来像主权财富基金在买赛博保险。
来源：Bloomberg | 2026-06-17