SemiAnalysis@SemiAnalysis_ · 6月23日43

This approach will mean that American robotics companies can't access the hardware they need for R&D.

译这种做法将意味着美国机器人公司无法获得研发所需的硬件。

Yuchen Jin@Yuchenj_UW · 6月23日26

Fable 5 is not back. Apparently Anthropic’s non-citizen researchers can’t use Mythos/Fable 5 due to the ban, but they can still build more powerful models like Mythos 6 or Fable 6. If that’s true, I don’t think the ban makes much sense.

译Fable 5 没有回归。 显然，由于禁令，Anthropic 的非公民研究员无法使用 Mythos/Fable 5，但他们仍然可以构建更强大的模型，如 Mythos 6 或 Fable 6。 如果真是这样，我认为这项禁令没什么意义。

Chubby♨️@kimmonismus · 6月22日23

Not again, @eucommission . Europe is once again being excluded from access to the latest SOTA scientific technology.

译Sakana AI发布Fugu Ultra，它不是传统意义上的前沿模型，而是一个学习编排层，将多个前沿模型转化为协调的智能体团队。其多智能体系统性能可与Fable 5相匹敌。然而，欧洲再次被欧盟委员会排除在访问这一最新SOTA技术之外，引发批评。

Chubby♨️@kimmonismus · 6月22日79

So Mythos "broke into almost all of NSA's classified systems, not in weeks, but in hours" and we already got the next iteration of Mythos. Big things are happening.

译据《经济学人》报道，6月11日，在亚马逊发现越狱的同一天，AI模型 Mythos 据称在数小时内攻破了美国国家安全局（NSA）和网络司令部的几乎所有机密系统，而非此前认为的数周。参议院情报委员会副主席 Mark Warner 转述 NSA 局长 Joshua Rudd 的话确认了此事。主推文还透露，Mythos 的下一轮迭代已经到来。

Rohan Paul@rohanpaul_ai · 6月22日65

The Economist: NSA leadership said that Anthropic’s Mythos helped break into “almost all” classified systems in hours, not weeks. The author also clarified that Mythos almost certainly worked with other tools under special test conditions. It is still alarming:

译经济学人报道，NSA 局长告知参议院情报委员会，Anthropic 的 AI 智能体 Mythos 在数小时内（而非数周）攻破了“几乎所有”机密系统。但文章作者 Shashank Joshi 澄清，该说法来自转述，且 Mythos 几乎肯定是在特殊测试条件下与其他工具协作完成的。他称原意是体现 Mythos 的强大能力，但未附加免责说明是个失误。

Berryxia.AI@berryxia · 6月21日57

讲真，AI时代并没有真正意义的平权可言！ 美国把Claude最强的Mythos模型限制出口后，居然还有大概200家机构能继续用。 换句话说：这200家美国机构现在手里握着的，是外面所有人都摸不到的顶级AI能力。 根据Bloomberg的报道，这些早期用户（包括一些银行、Cisco、Dragos等组织）在出口管制后依然保留着访问权限。 而普通用户和大多数组织已经被挡在外面。 评论区有人说得好："What can we peasant do?"（我们普通人能干嘛？）作者回了一个词："Permanent underclass"（永久底层阶级）， 说真的，这话虽然扎心但不无道理。 想想看，当200家顶尖机构已经在用最顶级的模型做研发的时候。 外面的人还在争论该用GPT-4还是Claude Sonnet，这个差距不是一两年能追回来的。 我感觉这才是真正的AI鸿沟，不是你买不买得起API的问题，是人家用的东西你连见都见不到。 对我们普通人来说，现在的策略就一个：先把现有工具用到极致，别等"更好的出来"，等你也等不到。

译美国对Anthropic最强模型Claude Mythos实施出口管制后，仍有约200家早期用户（包括部分银行、Cisco、Dragos等）保留访问权限，普通用户和大多数组织已被挡在门外。据Bloomberg报道，这200家机构手握外界无法触及的顶级AI能力，形成“永久底层阶级”式的AI鸿沟。差距不是简单API购买能力能弥补，普通用户甚至见不到这些模型。

Chubby♨️@kimmonismus · 6月20日72

Trump’s perspective on Anthropic as a possible national security risk, as well as his views on AI energy needs and regulatory policy. Let’s hope we’ll get fable 5 soon again.

译特朗普将Anthropic视为潜在国家安全风险的观点，以及他对AI能源需求和监管政策的看法。 让我们希望很快能再次看到fable 5。

Rohan Paul@rohanpaul_ai · 6月20日61

Not anymore: Trump on whether he sees Anthropic threatening national security Axios' Marc Caputo asked Trump: "Do you view Anthropic, and to a degree its CEO Dario Amodei, as a threat to national security?" President Trump: "Well, not now, but a week ago, maybe....We were in the room in the G7 and seems like a nice guy, smart guy." Source: Axios

译特朗普在Axios采访中被问及是否认为Anthropic及其CEO Dario Amodei威胁国家安全时回应“现在不，但一周前可能”。Axios报道称，特朗普政府一周前曾限制访问Anthropic的新模型，彼时他可能视其为威胁，并称“人们会立即因泄露被关进监狱”。

X.PIN@thexpin · 6月20日62

http://x.com/i/article/2067542907339628544 # How Chinese Tech Found a Way Around America's Restrictions Hong Kong is the operating system for Chinese tech companies that America no longer wants, but can't quite cut off. Montage Technology’s prospectus contained an unusual admission. The Shanghai-based chip designer had consulted its legal advisors, who determined it qualified as a “covered foreign person” under U.S. Treasury rules governing outbound investment into sensitive Chinese tech. The implications were pointed: for certain American investors, buying its shares could require notifying the federal government. Montage disclosed this on page one. Then it listed on the Hong Kong Stock Exchange anyway. Its cornerstone investor was J.P. Morgan. I’ve been following China’s semiconductor story for a while now, and the Montage case stopped me cold. Not because it was shocking, but because it was so perfectly ordinary. This is just how things work now. It also forced me to rethink what Hong Kong actually is. I used to think of it as a diminished version of its former self: less global, more Chinese, gradually losing the qualities that made it matter. But that framing misses something. Hong Kong isn’t becoming less useful. It’s becoming useful in a different way. It has evolved into something like an operating system for Chinese tech companies that America no longer wants but can’t quite cut off. The listings are the interface. The capital flows are the output. And the architecture underneath, largely invisible, is what makes the whole thing run. ## The Kind of Company Washington Worries About Montage doesn’t build AI chips. It builds something more mundane: the memory interface components without which AI servers cannot function. There are only three serious players in the world. Renesas of Japan. Rambus of the United States. And Montage, headquartered in Shanghai. That is enough to put it on Washington’s radar. Montage’s decision to list in Hong Kong wasn’t really a decision. It was the elimination of alternatives. Shanghai’s STAR Market suspended its listing channel for unprofitable tech firms in August 2023. A domestic listing would have meant profit thresholds Montage couldn’t meet, political scrutiny it didn’t need, and a renminbi raise that couldn’t fund overseas R&D or foreign hires. None of that works for a company trying to compete with Renesas and Rambus on a global stage. New York was simpler to rule out. I kept coming back to a line from Samuel Bickett, a former compliance director at Bank of America Merrill Lynch who spent years working in Hong Kong’s financial system and has testified before Congress on the subject. His read on the situation was direct: “A company that was big enough would previously have tried to list in New York. Now, a lot of them really can’t do that. It’s too risky.” The same Treasury rule that flagged Montage as a “covered foreign person” makes a U.S. listing a compliance nightmare before a single share changes hands. So Hong Kong. Not as a preference. As the only door still open. Montage is not alone. The more I looked at recent Chinese AI IPOs, the clearer the pattern became: almost all of them were ending up in Hong Kong. Twenty-three of twenty-seven Chinese AI-related companies that have gone public in 2026 listed there, a rate above 85%, according to Kharon’s analysis. The scale of the shift is hard to overstate: More than $14 billion was raised in the first quarter alone across 40 listings, a 489% increase from the same period last year — the strongest Q1 performance the exchange has seen in five years, according to KPMG’s official quarterly review. Analysts expect the full-year total to approach $60 billion. More than 400 companies are in the queue. Hong Kong’s Chapter 18C rules, introduced in 2023, allow pre-profit companies in AI, semiconductors, robotics, and autonomous driving to list on the main board without mandatory profit forecasts. The door that Beijing closed on the mainland, Hong Kong reopened offshore. If you enjoyed reading this, please subscribe! ## When the Entity List Becomes a Marketing Document If Montage represents the quiet end of this story, a chip company slipping through a legal gap with minimal fanfare, then Biren Technology represents the loud end. Biren makes GPUs. Not the kind that render video games, but the kind that train AI models: the direct category Washington has spent the most energy trying to keep out of Chinese hands. Founded in 2019, the Shanghai-based company built its BR100 chip as an explicit domestic alternative to Nvidia’s processors, targeting data centers, AI inference, and cloud computing workloads. It was, by design, exactly the kind of company the U.S. export control regime was meant to contain. In October 2023, Washington added Biren to the Entity List with a strict designation that complicated its access to overseas foundries and chip manufacturing equipment. Then it went public in Hong Kong on January 2, 2026, the exchange’s first listing of the year. Retail investors oversubscribed the offering 2,347 times. Institutional demand hit 26 times available shares. The stock opened at HK$35.70 against an IPO price of HK$19.60 and briefly surged 119% intraday. By the close of its first trading day, Biren was worth roughly $11 billion. The Entity List designation is there, in the prospectus, disclosed as a risk factor. Investors read it. They bought anyway. This is the detail I keep thinking about. Washington’s logic is that designating a company disrupts its access to capital, technology, and global markets. In Biren’s case, the designation appears to have done the opposite: it confirmed the company’s strategic importance, signaled it was a serious player in the domestic GPU race, and made it a more compelling story for investors betting on China’s self-sufficiency drive. The Entity List, intended as a wall, functioned instead as a validator. Biren couldn’t have achieved this on a mainland exchange. Its losses were widening, its revenues still nascent. Under normal STAR Market rules, the listing would have been impossible. Chapter 18C made it happen in Hong Kong. And the dollars it raised, not renminbi, dollars, will fund the next generation of chips that Washington is trying to prevent from being built. Most coverage of this IPO boom focuses on the listings. The more important story is the mechanism underneath them. China maintains strict capital controls. Foreign investors cannot easily buy shares on the Shanghai or Shenzhen exchanges directly. But through Stock Connect, the cross-border trading link between Hong Kong and the mainland exchanges, they can. A foreign investor with a Hong Kong brokerage account can buy A-shares listed in Shanghai or Shenzhen, and mainland investors can access Hong Kong stocks the same way. The exchange functions as a revolving door: not just a listing venue, but a bypass around China’s capital controls in both directions. This is what makes Hong Kong structurally irreplaceable for Beijing, and structurally inconvenient for Washington. As I was digging through research on Hong Kong’s role in global finance, one statistic kept resurfacing: roughly two-thirds of Chinese entities sanctioned under U.S. law still have securities accessible through Hong Kong’s markets, either through direct listings or through Stock Connect, according to a Hong Kong Watch report presented at the Hudson Institute. The implication is one that rarely gets stated plainly: you cannot import goods from these companies because of forced labor concerns, but you can invest in them. That cognitive dissonance is not accidental. It is built into the architecture of the system. Bickett offered what I think is the clearest description of what’s actually happening: “What we’re seeing in Hong Kong is a pretty clear shift from a global financial center to a very China-focused instrument for Chinese companies to interface with the world.” Continue Reading

译美国财政部对外投资限制下，上海芯片设计公司Montage被认定为“covered foreign person”，无法在美上市，也不满足A股盈利要求，最终选择港交所上市，基石投资者为摩根大通。Montage并非孤例：2026年迄今，27家中国AI相关上市公司中23家选择香港（超85%）；Q1港股40宗IPO融资超140亿美元，同比暴增489%，全年预计接近600亿美元，超400家企业在排队。香港2023年推出的18C章规则允许AI、半导体等未盈利公司在主板上市，成为北京关闭A股门前中国科技企业的唯一出口。

Berryxia.AI@berryxia · 6月20日51

亚马逊的骚操作简直了！ 自己是A社的竞争对手也是股东… Trump前几天还把Anthropic当成国家安全威胁，这两天刚跟Dario见完面就改口说“是个好人、聪明人，反应很负责任”。 记者问他是否认为Anthropic和Dario是国家安全威胁，Trump说：“现在不觉得了，但一周前可能觉得。” 他昨天跟Dario见了面，觉得对方反应很快、很负责任。谈到是否用国防生产法管制AI时，他说“可能会，但目前看来大家都很负责任”。 最后还补了一句：“其实是竞争对手兼部分股东把Anthropic举报的……” 这波操作把AI监管的真实逻辑给暴露得挺清楚的：出口管制这种级别的决定，一周前可能是威胁，一次见面后就变成了“聪明好人”。 而举报方据说是Amazon（既是竞争对手又是Anthropic的股东）。 以前大家觉得AI政策会越来越制度化、越来越基于技术评估，现在看起来至少在美国，还挺依赖“谁跟总统聊了、聊得怎么样”。

译特朗普对Anthropic的态度在一天内反转：此前将其视为国家安全威胁，但会见CEO Dario Amodei后改口称其“聪明、好人、反应很负责任”。被问是否用国防生产法管制AI时，他表示“可能但似乎大家都很负责任”。特朗普还透露，举报Anthropic的是其竞争对手兼部分股东——Amazon。这一事件暴露了AI监管的真实逻辑：出口管制等重大决策可能取决于与总统的面谈效果。

SemiAnalysis@SemiAnalysis_ · 6月20日56

Everyone's talking about Elon Musk's AI1 satellite this week. Almost nobody noticed: China moved on space-based AI compute a week BEFORE he did. Last week, Beijing quietly launched its first Space Computing Industry Innovation Center. Government-chartered, led by BUPT, a top state telecom university. The mandate: radiation-hardened space-native AI chips, compute satellite platforms, space-optimized LLMs, integrated space-ground networking, and "tokenized" orbital compute operations. 24 hours later, Beijing E-Town convened satellite makers, chip firms, and materials companies to plan a Space Intelligent Computing Research Institute.

译上周，北京低调启动首个太空计算产业创新中心，由政府特许、北京邮电大学（BUPT）主导。其任务涵盖抗辐射太空原生AI芯片、计算卫星平台、太空优化大语言模型、天地一体化网络及“tokenized”轨道计算运营。次日，北京亦庄又召集卫星制造商、芯片公司与材料企业，规划建设太空智能计算研究院。

Chubby♨️@kimmonismus · 6月20日80

Trump viewed Anthropic and CEO Dario Amodei as a threat to national security. „Well, not now, but a week ago, maybe," the president said.“ Tone changed. Now he thinks Dario is smart and a nice guy.

译特朗普在Axios专访中称Anthropic与其CEO Dario Amodei曾构成国家安全威胁，但随后态度转变，认为Dario既聪明又是个好人。引用推文证实了Axios的独家报道。特朗普表示“不是现在，但一周前也许是”，暗示其立场随时间变化。

Rohan Paul@rohanpaul_ai · 6月19日68

ASML just became the center of a US-China chip fight after Washington said it fears a banned EUV lithography tool may have reached China, while ASML says it never shipped one there. US Commerce Secretary Howard Lutnick outlined concerns to ASML’s senior leaders that one of its top-of-the-line machines may have made its way into China, in violation of US-led export restrictions. EUV uses 13.5nm light to print the smallest chip patterns, which lets foundries pack more transistors into AI processors and phones. ASML however, denies selling extreme ultraviolet chip tool to China. For context, ASML is the sole produce of EUV lithography machines, and each one costs roughly $250M today, with newer High-NA EUV versions closer to $ 400M. And ASML sold of only 48 EUV systems in 2025, and its market share in EUV is effectively 100%. There are 314 EUV machines in operation around the world, 26 that have been decommissioned — and none in China, according to Bloomberg. Senior US administration officials said that they have evidence of ASML shipping to China specialty equipment used to transport EUV machines, as well as other components that could be used in EUV systems. The proof of those exports is too sensitive to disclose, they said Reuters had already reported a Shenzhen EUV prototype built with former ASML engineers, but making EUV light is still far from running a factory-grade production tool. The hard part is repeatable nanometer precision across optics, masks, software, stages, wafers, maintenance, and defect control. Shares of ASML fell ~2.6% in morning trading after this report . China’s blocked access to EUV equipment may be Huawei Technologies Co’s biggest chipmaking hurdle, as it tries to rival Nvidia in AI chips. Huawei’s chip head recently highlighted how the company is building chips without ASML’s EUV machines. If any EUV tool reached China, it would likely rank as 1 of the biggest known breaches of US-led export controls meant to restrict Beijing’s access to military-useful AI. There are also gaps between what US and non-US companies can sell to certain Chinese customers on Washington’s entity list, which places stricter limits on named firms than on China overall. For Huawei, the US has basically blocked nearly all technology access and used rules beyond its borders to apply those limits to firms like ASML. But US rules still permit some overseas dealings with other listed Chinese companies, including Huawei partner SwaySure Technology Co, which joined the list in 2024. Washington does have the power to force ASML to fully cut ties with SwaySure, but President Donald Trump’s team has not used it. The Netherlands has already placed curbs on ASML selling EUV machines and some advanced immersion deep ultraviolet, or DUV, lithography tools to China. But the Biden administration grew frustrated when ASML increased shipments of gear facing upcoming bans before certain DUV controls were officially enforced. Now if indeed EUV export-control violation is proven, that would certainly not mean the US “bans ASML,” because ASML is too central to TSMC, Samsung, Intel, Nvidia, Apple, and US chip policy. The more realistic damage would be Dutch investigations, fines, forced compliance controls, tougher license rules, and pressure to cut ASML’s China service, spare parts, software updates, and DUV sales, which could hurt the roughly 20% of 2026 revenue ASML expects from China. --- reuters .com/world/china/us-tells-asml-it-is-concerned-china-may-have-top-chip-tool-bloomberg-news-2026-06-19/

译美国商务部长Lutnick向ASML高层提出担忧，指一台顶级EUV光刻机可能已流入中国，违反出口管制。ASML否认向中国出售EUV设备。EUV每台约2.5亿美元，全球运行314台，据彭博中国为零。若违规属实将是美国对华芯片管制重大突破。ASML股价早盘跌约2.6%。路透曾报道深圳用ASML前员工搭建原型机但离量产有差距。美国可强制ASML切断与中国客户合作，荷兰已限制EUV及部分DUV出口，进一步收紧可能影响ASML约20%的2026年中国区收入。

Nathan Lambert@natolambert · 6月19日52

It'll come down to, if the U.S. labs don't want distillation they shouldn't have an API. Seems like eventually they'll do this for some models, and that's their choice to make. More onerous regulation wont really work and will hurt startups in the US.

译归根结底，如果美国实验室不想被蒸馏，他们就不该提供API。看起来他们最终会对某些模型这么做，这是他们的选择。 更严格的监管实际上不会奏效，反而会伤害美国的初创公司。

Rohan Paul@rohanpaul_ai · 6月19日67

The White House and Anthropic may have found the first serious path to restore Mythos and Fable access without pretending jailbreaks can be eliminated. AI regulation may be shifting from vague fear to a benchmark based tests of model failure, because completely removing absolutely all jailbreak is probably not a possible target. The proposed framework would score how far the bypass went, what capabilities became reachable, how repeatable the attack was, and whether the exposed behavior had real operational consequences. Both sides are now moving toward a shared way to score what a jailbreak actually means. The hard truth is that perfect immunity is probably the wrong target; a recent red-team study found even hardened frontier models still produced confirmed harmful completions under sustained automated attack, with Fable 5 remaining more robust than Opus 4.8 but not invulnerable. So once, for a newly released model, the governments can ask the same questions every time, how much was bypassed, what capability was exposed, how reproducible was the attack, and what damage could follow, thats a much more practical path.

译白宫与Anthropic正合作制定正式技术评估框架，用于量化AI模型越狱攻击的严重性，并建立标准化评估方法论。框架将开发通用基准，评估防护被绕过程度、暴露的能力、攻击可重复性及实际运营后果。双方认识到完全免疫越狱并非可行目标。近期红队研究表明，强化后的前沿模型Fable 5比Opus 4.8更鲁棒，但在持续自动化攻击下仍产生有害输出。新框架每次以相同问题衡量风险，被视为更务实的监管路径。

Chubby♨️@kimmonismus · 6月19日52

Early Users of Anthropic Mythos still have access after US order. Mainly through project Glasswing. Via Bloomberg

译Anthropic Mythos 的早期用户在美国命令后仍可访问。 主要通过 Glasswing 项目。据 Bloomberg 报道。

Rohan Paul@rohanpaul_ai · 6月19日70

Nypost: Anthropic is trying to get Washington to reverse the US block on its most powerful Mythos Anthropic has proposed working more closely with the Trump administration, improving communication, and resolving security concerns faster as it seeks to end the US restrictions. --- nypost .com/2026/06/18/business/anthropic-floats-proposal-to-lutnick-to-end-us-ban-of-powerful-mythos-fable-ai-models-sources/

译《纽约邮报》：Anthropic正试图让华盛顿撤销对其最强大模型Mythos的美国禁令。 Anthropic提议与特朗普政府更紧密合作，改善沟通，更快解决安全问题，以结束美国限制。

Rohan Paul@rohanpaul_ai · 6月19日56

Anthropic vs. The US Government The kill-switch debate got user feedback.

译Anthropic 正试图让华盛顿撤销对其最强 AI 模型 Mythos 的出口限制。据纽约邮报报道，Anthropic 已向特朗普政府提交提案，承诺更紧密合作、改善沟通并更快解决安全关切，以换取解除禁令。与此同时，"杀开关"（kill-switch）讨论获得了用户反馈。

Chubby♨️@kimmonismus · 6月19日55

The third major win for OpenAI: Ha Thai has departed Meta to join OpenAI to lead communications for devices As Axios says: OpenAI is expected to unveil its first device this year. Therefore, they hired a key figure from a competitor to prepare the hardware release. Things have been rather quiet regarding the hardware device so far. But this makes it clear that OpenAI will continue down this path.

译OpenAI近期完成两起关键人事引进：从Meta挖来Ha Thai负责设备通讯，Axios称OpenAI预计今年推出首款设备；同时，曾帮助塑造特朗普政府政策的Dean Ball也加入OpenAI。在最大竞争对手Anthropic与美国政府关系紧张的背景下，OpenAI借由吸纳这些与政府关系密切的人物，提前应对潜在冲突并强化政府渠道。除开源外，OpenAI成为Anthropic当前困境的最大赢家。

Rohan Paul@rohanpaul_ai · 6月19日52

US approves faster grid connections for data centers. The plan lets PJM review up to 10 projects a year, but only projects of at least 250MW qualify, i.e. this is aimed at big power plants, not small local upgrades. The bottleneck is interconnection, the slow engineering and approval process that decides whether a new generator can safely plug into the grid without overloading wires, transformers, or protection systems. PJM says the threat is a possible 60GW shortfall over the next decade, and the pressure is coming from data centers that need steady electricity for servers, cooling, networking, and backup systems. The new track runs from July-26 through 2027, and selected projects are supposed to get grid agreements within 10 months and start operating within 3 years. --- wric .com/news/politics/ap-federal-regulators-back-trumps-plan-to-speed-power-to-energy-hungry-ai-data-centers/

译美国批准加快数据中心电网接入计划。PJM每年最多审查10个项目，但每个项目容量须至少250MW，主要针对大型发电厂。瓶颈在于互联审批流程缓慢。PJM警告未来10年可能出现60GW电力缺口，压力来自数据中心对服务器、冷却等持续供电需求。新计划从2026年7月运行至2027年，入选项目预计在10个月内获得电网协议，3年内投入运营。

Chubby♨️@kimmonismus · 6月18日47

Anthropics founder and co founder are working hard to get fable 5 back for everyone. Looking good, security issues are being addressed. Via Bloomberg

译Anthropic 的创始人和联合创始人正在努力让 Fable 5 重新为所有人可用。 看起来不错，安全问题正在解决。Via Bloomberg

X.PIN@thexpin · 6月18日47

Alibaba Cloud launched its first data centers in France this week, its third European hub after Germany and the UK. Agentic AI services are planned for the region in the second half of the year. The timing is pointed. The European Commission published its Cloud and AI Development Act on June 3rd, explicitly flagging "limited data center capacity" as a threat to European digital independence. The EU wants to be a global AI leader. It needs infrastructure to do it. Europe's data sovereignty push is largely about reducing dependence on US hyperscalers — AWS, Azure, Google Cloud. Whether European businesses and regulators will draw that distinction sharply remains to be seen .So far, Alibaba Cloud's European expansion has proceeded without significant political friction — Germany and the UK came first, France now.

译阿里云本周在法国启用了首个数据中心，这是继德国和英国之后的第三个欧洲节点，计划今年下半年在该地区推出智能体AI服务。此举正值欧盟委员会6月3日发布《云计算和AI发展法案》，该法案明确指出“有限的数据中心容量”威胁欧洲数字独立。欧洲的数据主权推动旨在减少对美国云巨头AWS、Azure、Google Cloud的依赖。目前阿里云在欧洲的扩张尚未遇到重大政治阻力。

X.PIN@thexpin · 6月18日61

China just opened its stock market to unprofitable AI companies. Shanghai's Star Market clarified rules this week: LLM developers can go public with zero revenue as long as they have a $591M+ market cap, a live model at scale, and a commercialization plan. Chinese AI labs- Moonshot, http://Z.ai, Manus, and others — have been burning cash at scale competing with each other and with US frontier labs. Venture funding exists but is unevenly distributed. Opening the public markets to pre-revenue AI companies is a structural attempt to broaden the capital base and keep Chinese labs in the race. The same rule update also expanded Star Market access for quantum technology, brain-computer interfaces, robotics, hydrogen energy, nuclear fusion, and 6G essentially a shopping list of strategic tech priorities dressed up as listing reform.

译中国科创板（Star Market）本周修订上市规则，允许未盈利AI公司上市。条件：市值不低于5.91亿美元、拥有规模化运行的模型并提交商业化计划。此举旨在为月之暗面（Moonshot）、Z.ai、Manus等持续烧钱的AI实验室拓宽融资渠道，维持其与海外前沿实验室的竞争。此外，同一规则更新还扩展至量子技术、脑机接口、机器人、氢能、核聚变和6G等战略性技术领域。

Rohan Paul@rohanpaul_ai · 6月18日52

Large AI campuses are indeed strategic infrastructure of a nation. US Department of Justice just moved to protect xAI’s Mississippi AI data center because the government sees high-speed AI infrastructure as part of America’s national security base. DOJ’s argument treats Grok and the systems around it as critical technology, not just consumer software, because advanced models can support defense operations. The move definitely reflects a wider policy shift: the U.S. does not want strategic AI workloads to depend on fragile supply chains, slow permitting cycles, or foreign-controlled compute capacity. AI compute is becoming a defense resource, where available power, data center scale, model capability, and operational readiness are now tightly linked.

译美国司法部近日保护 xAI 位于密西西比州的 AI 数据中心，认为高速 AI 基础设施属于美国国家安全基础。司法部将 Grok 及其系统视为关键技术而非普通消费软件，因先进模型可支持国防行动。此举反映政策转变：美国不希望战略 AI 工作负载依赖脆弱供应链、缓慢审批流程或外国控制算力。AI 算力正成为国防资源，可用电力、数据中心规模、模型能力与运营准备度紧密关联。

SemiAnalysis@SemiAnalysis_ · 6月18日40

POV: The G7 Deep State Committee is meeting in France this week to decide the future of Western AI. The summit is hosted in Évian-les-Bains, near the Swiss Alps and a 5-hour train ride from the Palace of Versailles.

译POV: G7深层政府委员会本周在法国开会决定西方AI的未来。峰会在埃维昂莱班举行，靠近瑞士阿尔卑斯山，距离凡尔赛宫5小时火车车程。

Berryxia.AI@berryxia · 6月18日59

卧槽，奥特曼和Dario “同台竞演”了！ G7峰会上，Sam Altman和Dario Amodei坐在同一桌吃饭，而美国刚刚限制了Anthropic最强模型的出口，这让盟友们有点尴尬。 在法国埃维昂的G7工作午餐上，OpenAI和Anthropic的两位CEO都出席了AI议题讨论。 但就在不久前，美国以安全为由对Anthropic的Mythos级模型实施了出口管制，直接影响了盟友获取最先进AI的能力。 这让现场氛围显得有些微妙。 以前AI发展主要是公司之间的技术竞赛，现在突然变成了国家层面的权力游戏。 美国一边用出口管制卡住前沿模型的扩散，一边让自家OpenAI的掌门人和盟友CEO一起吃饭谈合作。 Anthropic作为美国公司，却因为模型太强而被自己国家限制，这本身就很讽刺。 这波操作把AI从“全球共享的技术”重新定义成了“战略资源”。 盟友想要最强模型，却要看美国脸色。 美国则在用管制手段维持技术优势，同时拉拢盟友一起制定规则。 真正的张力不在于技术本身，而在于谁来决定谁能用最强的AI。 以前大家觉得AI监管主要是安全和伦理问题，现在G7这种场合直接把地缘政治和产业控制摆上了台面。 未来AI的发展，可能越来越不是“谁模型强”，而是“谁能控制模型的出口和使用权”。

译G7峰会上，OpenAI CEO Sam Altman与Anthropic CEO Dario Amodei同桌参加AI工作午餐。然而美国此前以安全为由对Anthropic的Mythos级模型实施出口管制，限制盟友获取最先进AI。现场氛围微妙，AI发展从公司技术竞赛变为国家权力游戏。美国一边用出口管制卡住前沿模型扩散，一边让自家CEO与盟友谈合作，Anthropic因模型过强被本国限制。此举重新定义AI为战略资源，盟友需看美国脸色，未来焦点将从模型能力转向控制权之争。

Chubby♨️@kimmonismus · 6月18日71

That sounds very bad for a soon-to-be re-release of Fable 5. "Trump administration officials tell WIRED that if Anthropic wants to rerelease Fable 5, it will need to ensure the model's guardrails can't be circumvented. Security experts say that can't be done." If this is true, we probably wont see a Fable 5 re-release any time soon.

译特朗普政府官员要求 Anthropic 在重新发布 Fable 5 前必须确保模型护栏无法被绕过，但安全专家认为这不可能实现。若该要求属实，Fable 5 短期内可能不会重新发布。

Rohan Paul@rohanpaul_ai · 6月18日72

The G7 AI lunch showed that frontier AI leaders have become part of the room where geopolitical choices are made. The meeting put Sam Altman, Dario Amodei, Demis Hassabis, Arthur Mensch, Aidan Gomez, Uljan Sharka, Victor Riparbelli, Robin Rombach, Alex Wang, Marc Benioff, and leaders from Sarvam and Sakana into the same room as national leaders. Sam Altman, who sat between President Donald Trump and Egyptian President Abdel Fattah el-Sisi, was the first CEO to speak at the hours-long lunch. The power shift is clear: governments can pass laws, but only a few private labs can build the models, test their dangerous abilities, restrict access, or provide the infrastructure needed to run them. In G7, the immediate fight is over frontier AI access, because Anthropic’s Fable 5 and Mythos 5 triggered U.S. export controls after officials worried advanced models could help find software weaknesses at scale. ---- From "Forbes" YouTube channel, (link in comment)

译G7 AI午餐会上，Sam Altman（坐于特朗普与埃及总统之间）、Dario Amodei、Demis Hassabis等前沿AI公司CEO与多国领导人同席，讨论地缘政治决策。Altman率先发言。推文指出权力转移：政府能立法，但只有少数私人实验室能构建模型、测试危险能力、限制访问或提供基础设施。当前G7焦点是前沿AI访问控制——Anthropic的Fable 5和Mythos 5触发了美国出口管制，因官员担忧这些先进模型可能大规模发现软件漏洞。

Nathan Lambert@natolambert · 6月18日69

We need a word for this sort of impossible demand with llms. It's almost like we've never dealt with a dual use technology before.

译特朗普政府官员要求Anthropic若重新发布模型Fable 5，必须确保其安全护栏无法被绕过（即实现“零越狱”）。安全专家则认为这在技术上不可行。主推文指出，这种对LLM的不切实际要求反映了双重用途技术面临的典型矛盾。

Rohan Paul@rohanpaul_ai · 6月18日68

Dario Amodei used the G7 meeting to argue that friendly nations need shared AI access, not separate rival systems. Sam Altman and Demis Hassabis backed the same broad idea, saying allies need shared cyberdefense tools, shared testing, and common technical standards. European and Indian leaders worried that if the US can “turn off the switch,” their economies and critical systems may become too dependent on US political decisions. --- ft .com/content/573925dd-6d41-4185-810d-2b848195903d

译Dario Amodei 在 G7 会议上提出，友好国家需要共享 AI 访问权限，而非各自为政的竞争系统。 Sam Altman 和 Demis Hassabis 支持这一总体思路，认为盟友需要共享网络防御工具、共享测试以及通用技术标准。 欧洲和印度领导人担心，如果美国能够“关掉开关”，他们的经济和关键系统可能过度依赖美国的政治决策。

Chubby♨️@kimmonismus · 6月18日79

We are entering a new era of the Cold War. Dario Amodei and Demis Hassabis are calling for a "U.S.-led coalition to shape rules and standards around artificial intelligence," excluding China. "Dario Amodei also said in his address that the coalition should structure access to frontier models and hardware - including both chips and other critical components - in a way that excludes China." The new Cold War will be a high-tech one in which the competition will be fundamentally excluded from all participation and involvement, because the technology affects national security and strategy.

译Dario Amodei（Anthropic）与Demis Hassabis（Google DeepMind）在G7闭门会议上呼吁组建美国主导的联盟，为人工智能制定全球规则和标准。Amodei指出，该联盟应以前沿模型和硬件（包括芯片及其他关键组件）的访问权限为手段，将中国排除在外。这一主张被评论为高技术新冷战的开端，竞争方将从根本上被剥夺参与权。

Chubby♨️@kimmonismus · 6月18日72

Although Trump says that developments regarding Anthropic are going well, the seating arrangement alone is remarkable. Via WSJ "At the lunch, OpenAI Chief Executive Sam Altman and Demis Hassabis, the head of Google’s DeepMind AI lab, were seated next to Trump. Amodei was seated on the opposite side of the table from the president, next to French President Emmanuel Macron." Crypto and AI expert David Sacks has just emphasized again in his last post how difficult working with Anthropic is and that he blames this: "Anthropic’s needlessly confrontational posture toward the Administration has distracted from that mission"

译特朗普与AI CEO午餐会，OpenAI的Sam Altman和Google DeepMind的Demis Hassabis被安排坐在特朗普两侧，而Anthropic的Dario Amodei则坐在对面、紧挨法国总统马克龙。尽管特朗普称Anthropic进展顺利，但座位安排已引关注。加密与AI专家David Sacks再次强调与Anthropic合作困难，指责其“对政府的好斗姿态分散了其使命”。

Chubby♨️@kimmonismus · 6月18日65

Very interesting: The NYT obtained and reported on internal Anthropic messages. The message: Employees feel "unfairly targeted" and "feel like they (the US government) don't want us to exist." However, the report omits the fact that Anthropic itself has repeatedly emphasized the security risks its model poses. Messages here:

译纽约时报获得并报道了Anthropic内部聊天记录，显示员工感到“被不公平针对”，认为美国政府“不希望我们存在”。但报道忽略了Anthropic此前曾反复强调自身模型安全风险的事实。引用推文指出，这些员工正是此前声称有能力带来网络安全“清算”的同一批人，如今却怀疑自己被政府“找茬、霸凌、不公平针对”，并向媒体泄露内部聊天。

Chubby♨️@kimmonismus · 6月17日55

Trump and the AI ​​CEOs are currently meeting. Judging by his facial expression, Sam seems to be enjoying himself more than Dario.

译特朗普与AI CEOs目前正在会面。从他的表情来看，Sam似乎比Dario更享受其中。

Chubby♨️@kimmonismus · 6月17日14

The silence emanating from Washington, which emerges from the talks between Anthropic and US officials, is far more worrying than any loud dispute.

译来自华盛顿的沉默，源自Anthropic与美国官员的会谈，比任何激烈的争论都更令人担忧。

Chubby♨️@kimmonismus · 6月17日71

AI labs are built on foreign talent. Thats a fact. Now the US is reportedly testing restrictions on "foreign persons" accessing frontier models. "The Trump administration appears to have targeted only Anthropic so far, warning the company on Friday in a letter from Commerce Secretary Howard Lutnick that it would need a license to make its latest models available to “foreign persons,” including its own employees. But Anthropic’s biggest rival, OpenAI, has flagged its concerns about the issue." (The Information) 38% of researchers publishing at leading AI conferences in 2024 got their undergrad education in China, per MacroPolo estimates cited by The Information. If US policy starts restricting model access by nationality, frontier labs are suddenly in a very difficult situation. That is why the issue surrounding Anthropic and Fable 5 is such a significant development, and that is why so much depends on the decisions made in the next few days.

译美国据报正考虑限制“外国人员”访问前沿AI模型。特朗普政府已首先针对Anthropic，要求其向包括自家员工在内的外国人员提供最新模型前必须取得许可证。OpenAI也表达担忧。据MacroPolo估算，2024年顶级AI会议论文作者中38%本科毕业于中国。若政策全面推行，前沿AI实验室将面临严峻处境。Anthropic与Fable 5相关事件因此意义重大。

Chubby♨️@kimmonismus · 6月17日93

Interesting: The US Commerce Secretary reportedly told Anthropic it needs government permission to export Fable 5 and Mythos 5 anywhere in the world. Even to “any foreign national regardless of location.” "Lutnick threatened Anthropic with criminal and civil penalties if it failed to comply, according to a copy of the directive obtained by Bloomberg News." Anthropic’s response: disable both models for all users. Two points to consider: 1) Even friendly nations are excluded from the most significant models. SOTA is increasingly becoming a strategic weapon, kept secret. 2) As a result, AI is now being shifted away from private companies and directly into government control.

译美国商务部长Howard Lutnick告知Anthropic，出口其最新模型Fable 5和Mythos 5至全球任何地方均需政府许可，甚至禁止向任何国籍的海外人士提供。Lutnick在信件中威胁，若不遵守将面临刑事和民事处罚。Anthropic回应：禁用这两个模型。推文作者指出，连友好国家也被排除在SOTA模型之外，AI正从私人公司转向政府控制。该信件全文由Bloomberg公开。

Rohan Paul@rohanpaul_ai · 6月17日82

The US just refused to give G7 allies special access to Anthropic’s Mythos 5 and Fable 5. Washington says a jailbreak may let users bypass Fable 5’s safety layer and reach dangerous vulnerability-finding behavior, while Anthropic says the issue is narrow and not unique to its models. The UK wanted a carve-out for British users, but US officials signaled that ally-by-ally exemptions make little sense if the risk is tied to the model itself. --- firstpost .com/tech/us-refuses-to-ease-restrictions-on-anthropics-mythos-5-and-fable-5-for-g7-nations-14023216.html

译美国拒绝向G7盟国提供Anthropic的Mythos 5和Fable 5模型的特殊访问权限。华盛顿方面称，存在jailbreak可能绕过Fable 5的安全层，导致用户触及危险漏洞发现行为。Anthropic回应称该问题范围窄且并非其模型独有。英国曾希望为英国用户争取豁免，但美方官员表示，若风险与模型本身相关，逐盟国豁免并无意义。

Chubby♨️@kimmonismus · 6月17日68

UK Prime Minister Keir Starmer reportedly asked the Trump administration for a carve-out so British nationals and companies could regain access to Anthropic’s most advanced AI models. The answer was no. A Trump administration official told The NY-Post that giving even a G7 ally an exemption would be “completely illogical.” This is no longer just about whether Fable or Mythos can be jailbroken. It is about whether frontier AI models are now being treated like strategic national-security assets. Even America’s closest allies may not get access if Washington decides the model is too sensitive. The US is still negotiating directly with Anthropic and Dario Amodei over how to resolve the ban. But the message to the rest of the world is already clear: access to the frontier may no longer be a commercial question. It may become a geopolitical privilege.

译据报道，英国首相Keir Starmer请求特朗普政府给予豁免，允许英国国民和公司重新访问Anthropic最先进的AI模型Mythos和Fable，但遭拒绝。白宫官员表示，即使对G7盟友给予豁免也“完全不合逻辑”，并称“不能让前沿模型失控”。这标志着前沿AI模型正被视作战略国家安全资产，即使是美国最亲密的盟友也可能无法获得。美国仍在与Anthropic和Dario Amodei直接谈判解决禁令，但对外释放信号：前沿模型访问权可能从商业问题变为地缘政治特权。

Chubby♨️@kimmonismus · 6月16日61

Update on Fable5/Anthropic: Anthropic flew its top security people to DC. The export controls are still there. Via Wired Anthropic and the Trump administration wrapped up talks on Monday with no resolution - the export controls on Claude Fable 5 are still in place. No end in sight. The company has spent days arguing that Washington's concerns about the model are overblown, a line it repeated to government researchers at the Commerce Department. Commerce Secretary Howard Lutnick joined by phone from the G7 summit in Evian, France, while cyber director Sean Cairncross sat the meeting out. On Anthropic's side, cofounder Tom Brown and external affairs head Sarah Heck led the discussions, with red-teaming chief Logan Graham and security researcher Nicholas Carlini flying to DC for the talks. The core disagreement is whether Fable 5's guardrails can be stripped away to unlock the more powerful Mythos capabilities underneath, the NSA thinks yes, Anthropic thinks the risk is overstated. For now there's no clear sense of what the next steps look like.

译Anthropic 与特朗普政府就 Claude Fable 5 出口管制谈判周一结束，无果而终。核心分歧：Fable 5 的护栏能否被移除以解锁更强大的 Mythos 能力——NSA 认为可以，Anthropic 否认风险。Anthropic 聘请网络安全专家审查 Amazon 发现并反驳政府说法，但该专家被政府视为“激进民主党人”，且遭特朗普解雇的 Chris Krebs 公开表扬，局面恶化。目前 Anthropic 正与商务部、CIA 及白宫科学顾问会谈合规问题，技术越狱风险已退居次要，公司正持续失去支持。