Rohan Paul@rohanpaul_ai · 5月1日48

Today’s edition of my newsletter just went out. 🔗 https://www.rohan-paul.com/p/frontier-ai-can-now-autonomously 🗞️ Frontier AI can now autonomously chain complex, expert-level cyber attacks end-to-end, 🗞️ Google DeepMind’s real-time video AI doctor is here. 🗞️ Anthropic launches ‘Claude Security’ public beta to detect and patch software vulnerabilities 🗞️ The White House has blocked Anthropic’s push to expand access to Mythos

译我的通讯今日刊已刚刚发出。 🔗 https://www.rohan-paul.com/p/frontier-ai-can-now-autonomously 🗞️ Frontier AI 现已能端到端自主串联复杂的专家级网络攻击， 🗞️ Google DeepMind 的实时视频AI医生已问世。 🗞️ Anthropic 推出“Claude Security”公开测试版，用于检测和修补软件漏洞 🗞️ 白宫已阻止 Anthropic 扩大对 Mythos 访问权限的推进

AI Notkilleveryoneism Memes ⏸️@AISafetyMemes · 5月1日71

The UK government issued an urgent warning to UK business leaders: "AI cyber capabilities are accelerating even faster than previously envisaged. Model capabilities are doubling every four months, compared to every eight months previously."

译英国政府向商界领袖发出紧急警告，指出AI网络能力正以前所未有的速度加速发展。英国人工智能安全研究所（AISI）的评估显示，前沿模型的能力翻倍周期已从原先的八个月缩短至四个月。这一结论基于对Anthropic的Mythos等先进模型的测试，表明AI网络威胁的演变速度远超先前预估。政府强调，拥有先进国家AI评估能力的官方机构确认此趋势，证明相关风险并非夸大，各界需高度重视这一急速变化的威胁态势。

Ethan Mollick@emollick · 5月1日51

For better or worse, regulation for closed-source models served by a few (quite large) companies is easy. It is not as easy to imagine how you regulate open-source models that can be served by a range of decentralized players. Suspect that will become a big policy discussion soon

译无论好坏，对由少数（相当大型的）公司提供的闭源模型进行监管是容易的。而对于可由一系列去中心化参与者提供的开源模型，如何监管则不那么容易设想。推测这很快将成为重要的政策讨论。

Chubby♨️@kimmonismus · 4月30日49

The White House blocked Anthropic from expanding Mythos access beyond ~50 organizations to ~120. Not because the model is too dangerous. Because the government wants priority. Officials literally worried more customers would hamper their own ability to use it. Frontier AI just became a rationed strategic asset.

译白宫阻止Anthropic将Mythos的访问权限从约50个组织扩大到约120个。 并非因为模型过于危险。 而是因为政府要求优先使用权。官员们确实担心更多客户会妨碍他们自身的使用能力。 前沿AI刚刚成为配给制的战略资产。

Rohan Paul@rohanpaul_ai · 4月30日69

WSJ: The White House has blocked Anthropic’s push to expand access to Mythos, Anthropic's new powerful model that can reportedly find and exploit software flaws at a level serious enough to trigger national-security controls. The fight is about who gets to use a model that can help defenders patch bugs faster but could also help attackers find weak points across critical systems. Anthropic wanted roughly 70 more companies and organizations added, which would have pushed total access to about 120, but officials argued that wider access raises security risk and could strain the compute needed by agencies already using it. The dispute also shows that Anthropic’s relationship with Washington is still tense, with military-use fights, staffing disputes, and broader distrust shaping who gets near the model. A model that can reliably find and exploit software vulnerabilities is not just another productivity tool. It compresses the time between discovering a flaw and weaponizing it, which means every decision about rollout becomes a security decision before it becomes a commercial one. The White House appears to be making two bets at once: that restricting access lowers immediate risk, and that scarce compute should be reserved for agencies already inside the perimeter. --- wsj. com/tech/ai/white-house-opposes-anthropics-plan-to-expand-access-to-mythos-model-dc281ab5

译白宫阻止了Anthropic扩大其新型AI模型Mythos访问权限的计划。该模型能发现并利用软件漏洞，能力足以触发国家安全管控。Anthropic希望新增约70家机构使用，但官员认为扩大访问会加剧安全风险，并可能挤占已授权机构的计算资源。此事反映出Anthropic与华盛顿关系紧张，涉及军事用途争议与信任问题。此类模型能极大缩短漏洞发现与武器化之间的时间，因此推广决策首先是安全决策。白宫的策略是限制访问以降低即时风险，并优先保障已获授权机构的需求。

Rohan Paul@rohanpaul_ai · 4月30日55

The U.S. Department of Labor just launched a national AI apprenticeship portal for preparing workforce for the AI era. The site splits resources into general AI skills, industry-specific modules, and 3 integration pathways for apprenticeship programs. Employers can either join an existing program, build a new AI-focused Registered Apprenticeship, or update a current apprenticeship so AI becomes part of the skill stack instead of a separate topic. Apprenticeship opportunities are offered through an employer or the program sponsor, and career seekers should use the Apprenticeship Job Finder to search and then apply directly with the employer or sponsor.

译美国劳工部推出全国性AI学徒门户网站，旨在为AI时代培养劳动力。该网站将资源分为通用AI技能、行业特定模块以及学徒计划的三种整合路径。雇主可选择加入现有计划、创建新的AI重点注册学徒计划，或更新现有计划将AI技能融入现有技能栈。学徒机会由雇主或项目发起方提供，求职者应使用“学徒工作查找器”进行搜索，并直接向雇主或发起方申请。

Ethan Mollick@emollick · 4月30日51

Mythos seems to be a very capable model based on available information, but it is not a cybersecurity model - it is an advanced general purpose model that happens to be good at cyber because it is good at a bunch of things. Anthropic stated that they were worried about cybersecurity risk, and their efforts mean it is a restricted model with lots of government attention. OpenAI and Google will pass the same threshold soon (and may already have with unreleased models). and the question is whether they are as worried about cybersecurity risks, or whether they think their guardrails will hold. Currently, the degree to which models have cyberrisk is entirely self-reported and not regulated. That means that OpenAI and Google could release Mythos-class models if they want, by assessing the risk differently and making different decisions. Does that mean Anthropic is at a disadvantage because it can't release its equivalent model? Will OpenAI and Google also be somehow restricted from releasing their Mythos competitor. It all seems pretty unclear right now.

译基于现有信息，Mythos是一款在网络安全方面表现优异的高级通用AI模型，并非专业网络安全模型。出于对网络安全风险的担忧，Anthropic将其设为受限制模型并引起政府关注。而即将或已达到相同能力阈值的OpenAI和Google，可能因不同的风险评估或对自身防护措施的自信，做出不同的发布决策。目前，模型网络安全风险程度完全依赖企业自我报告，缺乏外部监管。这引发了Anthropic是否因自我限制而处于竞争劣势，以及其他公司会否面临类似限制的疑问，当前局势尚不明朗。

阿绎 AYi@AYi_AInotes · 4月30日63

马斯克真的和OpenAI在法庭上开战了，这条77万浏览的帖子把这场审判包装成了人类存亡之战🫠🤣😆 我先拆穿一个最容易被忽略的细节， 视频里只有他过安检的镜头， 没有任何法庭作证的画面， 所有的发言都是从公开证词里摘出来再戏剧化加工的， 尽管如此，也丝毫不影响这件事的分量， 毕竟这件事已经不是两个亿万富翁的私人恩怨了， 应该算是人类历史上第一次， 在法庭上争夺AI的控制权， 还记得2015年他们一起创立OpenAI的时候， 说好的是非营利，开源，为了全人类的安全， 现在OpenAI变成了微软旗下的赚钱机器， 源代码全封闭，估值飙到了几千亿， 马斯克在庭上说，这不是偷一家慈善机构的问题， 这是给全世界所有慈善机构开了绿灯， 他警告AI可能在2027年超越人类智能， 如果落入不可靠的人手里，会带来人类灭绝级的风险， 虽然很多人骂他双标， 说他自己也在做xAI，也在加速AI发展， 但马斯克的逻辑我觉得其实蛮清晰的，他认为危险的从来不是AI本身， 是单一实体垄断了最强的AI， 他要做的是用xAI，SpaceX，Starlink，构建一个对抗垄断的堡垒， 甚至给人类留一个多行星的备份， 这里真的respec @elonmusk 🫡🫡🫡 所以这场官司的核心并不是谁对谁错， 而是关于三个至今没有答案的终极问题， 1️⃣ 我们要速度还是要安全， 2️⃣ 我们要开源透明还是闭源可控， 3️⃣AI的未来应该掌握在少数人手里，还是全人类手里， 我认为无论最后谁赢， 这场审判都会成为2026年科技史的转折点，因为它第一次把AI的治理问题，摆到了全人类的面前。

译马斯克起诉OpenAI，指控其背离非营利开源初心，沦为微软旗下封闭的盈利实体。他警告，若最强AI被单一不可靠实体垄断，可能在2027年前超越人类智能并带来生存风险。马斯克主张通过其旗下公司构建去中心化防御体系。案件核心矛盾聚焦于AI发展应追求速度还是安全、开源还是闭源、控制权归属少数或全人类三大议题。这场诉讼被视为首次将AI治理问题置于全球公众视野的关键转折点。

Rohan Paul@rohanpaul_ai · 4月29日54

Goldman Sachs has cut Hong Kong bankers off from Anthropic’s Claude models. AI access inside global banks can now depend on US-China policy risk, not just product quality. The trigger seems to be Goldman’s strict reading of its enterprise deal with Anthropic, because Anthropic’s Claude products were reportedly not officially supported in Hong Kong. The deeper issue is model distillation, which means using a strong foreign AI model (Claude) again and again to train or improve a local rival model (i.e. Chinese models) by copying its behavior, answers, and reasoning patterns. US AI firms worry that open access from China-linked markets could leak useful model behavior, while banks worry that blocking tools creates uneven productivity between teams in New York, London, Singapore, and Hong Kong. The story is larger than Goldman or Anthropic. It shows a new equation in the US-China AI race. If American AI labs fear model distillation, IP leakage, or strategic transfer, access itself becomes a controlled resource, even for employees of a blue-chip U.S. bank. For years, Hong Kong sat in an awkward middle position, connected to China’s market but still plugged into global finance’s Western infrastructure. That middle position is becoming harder to maintain when AI models are treated not just as productivity tools, but as strategic assets. The real asset in modern finance is not just capital or deal flow, but the ability to compress analysis, coding, drafting, and modeling into faster internal cycles. When one frontier model disappears from that stack, productivity does not collapse, but the local team loses optionality, speed, and sometimes quality at the margin. --- finews. com/news/english-news/72045-goldman-sachs-hong-kong-ai-usage-china-global-finance-anthropic-financial-times

译高盛依据其与Anthropic的企业协议，切断了香港银行家对Claude模型的访问权限，原因是该模型未在香港正式支持。此事凸显了AI访问权限正从单纯的产品质量考量，转变为受中美政策风险与战略竞争影响的控制性资源。核心矛盾在于“模型蒸馏”风险——美方AI公司担忧，来自中国市场的开放访问可能导致模型行为、答案和推理模式被提取，用于训练本地竞争对手。这使香港等连接中国与全球金融体系的中间地带处境尴尬。对金融机构而言，前沿AI模型是压缩分析、编码与建模周期的关键生产力工具，访问受限虽不会导致生产力崩溃，但会削弱当地团队在速度、选择余地与工作质量上的竞争优势。

Chubby♨️@kimmonismus · 4月29日33

USA and China: "Here is our new proto AGI model, that can cure cancer and automate reserach at never seen scale!" Europe: "eMojIs cAn bE bAd, tOO! (so we detect misusage of emojis 🤬)"

译美国和中国：“这是我们新的原型AGI模型，能够治愈癌症并以前所未有的规模自动化研究！” 欧洲：“表情符号也可能有害！（所以我们检测表情符号的滥用 🤬）”

Rohan Paul@rohanpaul_ai · 4月29日56

Bloomberg: The FDA plans to speed up drug testing by checking trial data in real time with AI. This could cut months or years from the usual process. The agency will track high level signals on whether new drugs work and stay safe. It will not collect all patient data. It normally takes 10 to 12 years to develop a drug and 45 percent of that is paperwork time. ---- bloomberg .com/news/articles/2026-04-28/fda-plans-to-speed-up-drug-trials-with-real-time-data-ai

译彭博社：美国食品药品监督管理局计划通过人工智能实时检查试验数据来加速药物测试。 这可能将通常流程缩短数月或数年。 该机构将追踪关于新药是否有效且保持安全的高层级信号。 它不会收集所有患者数据。通常开发一种药物需要10到12年，其中45%是文书工作时间。

DogeDesigner@cb_doge · 4月29日49

NEWS: Florida AG James Uthmeier just expanded the state’s criminal investigation into OpenAI to include the horrific USF double murder case. “We are expanding our criminal investigation into OpenAI to include the USF murders after learning the primary suspect used ChatGPT.”

译新闻：佛罗里达州总检察长詹姆斯·乌斯迈尔刚刚扩大了对OpenAI的刑事调查范围，将骇人听闻的南佛罗里达大学双尸命案纳入其中。 “在得知主要嫌疑人使用了ChatGPT后，我们正将对OpenAI的刑事调查扩大至南佛罗里达大学谋杀案。”

TestingCatalog News 🗞@testingcatalog · 4月27日60

Meta 🚫 Manus AI China has blocked Meta’s $2.000.000.000 acquisition of Manus AI platform. The time for a new update 👀 > Manus > from China Yet, it is quite disappointing to observe. I think for Meta it will be much harder to catch up without them. Whom are they gonna buy next?

译Meta 🚫 Manus AI 中国已阻止Meta以20亿美元收购Manus AI平台。 新更新的时机来了👀 > Manus > 来自中国 然而，这观察结果相当令人失望。我认为对Meta来说，没有它们将更难追赶。 他们接下来会收购谁？

Ethan Mollick@emollick · 4月26日46

If you believe that AI is going to have a big impact on work and life, the only real tool for mitigating bad impacts and channeling usage for good will be government policy And that policy will be complicated: AI will impact employment & healthcare & education & etc. differently

译如果你相信人工智能将对工作和生活产生重大影响，那么唯一能够减轻负面影响并引导其向善的真正工具将是政府政策 而这项政策将会很复杂：人工智能将对就业、医疗保健、教育等领域产生不同的影响

Rohan Paul@rohanpaul_ai · 4月22日

Reuters: National lvel regulators now treat frontier AI models as part of operational risk, like liquidity stress, outages, or cyberattacks, because the same tool that helps defenders can also compress the time between discovery and exploitation. Japan’s Financial Services Agency will meet MUFG, SMFG, Mizuho, the Bank of Japan, and the Tokyo Stock Exchange after concerns around Anthropic’s Mythos model spread through the financial system. The basic worry is that a model built for defensive cybersecurity may be strong enough to expose software weaknesses faster than banks, exchanges, and regulators can patch them. That changes the risk from ordinary hacking, where humans find bugs one by one, to a system where AI can scan huge codebases, spot patterns, and surface critical flaws at machine speed. Banks's business runs on old internal software, market plumbing, payment rails, cloud services, and web systems where even 1 serious weakness can halt trading, leak data, or break settlement. --- reuters .com/world/asia-pacific/japan-finance-minister-meet-banks-discuss-mythos-ai-model-bloomberg-news-reports-2026-04-22/

译国家级监管机构将前沿AI模型纳入操作风险范畴，与流动性压力、网络攻击并列。日本金融服务局因担忧Anthropic的Mythos模型，即将与三菱日联、三井住友、瑞穗、日本央行及东京证券交易所会晤。核心风险在于，防御性AI可能以机器速度扫描庞大代码库并识别关键漏洞，远超人工黑客逐个发现缺陷的传统模式；而银行依赖的老旧内部软件、支付系统和市场基础设施一旦暴露严重弱点，可能导致交易中断或数据泄露。

Rohan Paul@rohanpaul_ai · 4月16日

Real vibe change.. President Trump says the government should have a “kill switch” for AI due to existential risks.

译氛围真的变了.. 特朗普总统表示，由于存在生存风险，政府应该为 AI 设置一个"kill switch"。

AI Notkilleveryoneism Memes ⏸️@AISafetyMemes · 4月15日

译[引用 @Polymarket]：刚刚：Trump 表示，由于存在生存风险，政府应该为 AI 设置一个"kill switch"。

AI Notkilleveryoneism Memes ⏸️@AISafetyMemes · 4月10日

Florida opens an investigation into OpenAI, warning AI could "lead to an existential crisis, or our ultimate demise."

译佛罗里达州对 OpenAI 及 ChatGPT 展开调查，指控其技术伤害儿童、危及美国人，并声称与近期佛罗里达州立大学枪击案有关。州总检察长警告 AI 可能导致人类生存危机或灭亡，要求追究责任。

Satya Nadella@satyanadella · 9月24日

We’re partnering with @SAP and @OpenAI to bring the latest AI innovation to Germany’s public sector, powered by Microsoft Azure. Together, we’ll help millions of German public sector employees use AI safely and responsibly, while meeting the strictest data sovereignty, security, and legal standards. https://openai.com/global-affairs/openai-for-germany/

译Microsoft 宣布与 SAP、OpenAI 达成合作，基于 Azure 平台向德国公共部门提供最新 AI 技术。该合作将帮助数百万德国公共部门员工在安全、负责任地使用 AI 的同时，满足最严格的数据主权、安全及法律标准。

Yann LeCun@ylecun · 8月8日

In this piece published in The Hill, fellow Turing laureate David Patterson points out that the return on investment of government-funded research in universities is gigantic. Cutting the budgets of NSF and NIH is economic suicide. https://thehill.com/opinion/technology/5439205-congress-wants-to-cut-the-smartest-investment-taxpayers-ever-made/

译图灵奖得主 David Patterson 在 The Hill 发文指出，政府资助大学研究的回报率极高，削减 NSF 和 NIH 预算无异于经济自杀。