内容来源：https://aiweekly.co/issues/the-ftc-wants-openais-and-anthropics-safety-incident-records

内容总结：

美国联邦贸易委员会（FTC）已向OpenAI和Anthropic分别发出50页民事调查令，要求提供安全事件、防护措施、营销材料和隐私政策相关记录。此类调查令类似传票，意味着正式调查启动，但并未认定两家公司违法。与此同时，Meta新商业条款将于10月30日生效，允许其抓取并利用出版商网站内容训练AI产品。据三名法律专家为Nieman Lab审阅，几乎所有新闻机构都拥有Meta商业账户，只要继续使用其商业工具即视为同意。Anthropic投资者日在讨论11月股份出售时提及约1.5万亿美元估值，低于此前约2万亿美元的水平；公司第三季度营收增长45%至168亿美元，增速较第二季度放缓。Character.ai已在特拉华州申请第11章破产保护，称平台开发和增长成本高昂，诉讼与监管成本使其无力产生足够现金流，迪士尼以2000万美元作为初始竞标方寻求授权部分知识产权。Citizen Lab发现，2024至2025年间Al Jazeera的45台设备通过在线广告遭受零点击攻击，广告来自以色列创立的营销公司Discover-Tech，携带被归因于Candiru的间谍软件，谷歌已封禁该公司广告平台。埃森哲自10月1日起仅按每人每月29美元集中支付Claude使用费用，此前部分员工月支出高达1000美元，超额部分需由项目预算承担。美国银行首席执行官布莱恩·莫伊尼汉向CNBC表示，该行在第二季度增聘1000名合同工，修补包括Anthropic的Mythos在内的AI模型发现的软件漏洞，相关修复与反洗钱合规成本在第三季度达“数亿美元以上”。此外，Meta发布8160亿参数的Muse Spark Open模型，可在苹果电脑本地运行，需至少1TB内存，最佳配置为搭载M5 Ultra芯片的Mac Studio；英伟达发布1200亿参数的Nemotron 3.5 Super VL开放模型，可处理PDF、图表、表格、图像和网页，并推出面向本地编码与研究的24GB模型Bolt。

中文翻译：

美国联邦贸易委员会（FTC）向OpenAI和Anthropic发出50页的记录调取要求，涉及安全事件、防护措施、营销和隐私方面，Axios报道。Meta的新商业条款于10月30日生效，允许其抓取并训练发布者网站的内容。Anthropic投资者日在场的投资者讨论了1.5万亿美元的IPO估值，较此前约2万亿美元有所下调。Character.ai申请破产。Citizen Lab追踪到通过定向广告向半岛电视台设备发送的间谍软件。埃森哲现在为每位员工每月支付29美元的Claude费用，美国银行则雇用了1，000名合同工来修补AI模型发现的漏洞。焦点在于：最大的开放模型现在可以在一张办公桌上运行。

赞助商
面向5人、10人或20人企业的企业级AI。无最低席位要求。无需CTO。
企业AI方案通常从20个席位起步，且需要技术人员来运维。Faradex可在几分钟内为你搭建专属的隔离AI服务器。符合条件的企业可享受两周免费试用，并获赠150美元AI额度。

行业动态
AI当下最热门的趋势，从应用榜单到社区信息流。完整汇总见《行业动态》。

• 一份订阅现在涵盖Grok、Cursor和X Premium。我们追踪的某个AI Telegram频道中，一条关于该捆绑套餐的帖子浏览量超过21，000次。xAI的X Subscription将Grok、Grok Bot、Cursor和X Premium整合到一个方案中，Grok应用与Cursor共享同一个Grok模型使用额度池。现有订阅者可切换并获得按比例退款。
• 亚马逊的Bee推出两款新的常听录音设备。Bee Frontier售价74.99美元，Bee Eclipse可夹在手机背面，售价99.99美元。两款设备均可免费总结你的对话，并保留14天记忆。Bee Pro方案为每月11.99美元。
• Bluesky用户现在可以通过描述来构建小型应用。Bluesky扩展了其Attie应用，让任何人都能制作网站、测验、投票和游戏，并与账户共存于同一网络。CEO Toni Schneider称之为“社交氛围编程”。
• 一款韩国AI角色聊天应用正在美国榜单上攀升。来自首尔Scatter Lab的zeta本周在美国娱乐榜上上升。zeta让用户免费与AI角色聊天，或在大约一分钟内创建自己的角色，并将聊天场景转化为图像。
• “10天预报”在美国谷歌上成为热门搜索。这与《自然》杂志一篇关于更长时效预报的新论文相呼应。一种机器学习偏差校正方法将ECMWF AI预报系统的2至6周预报能力提高了一倍，并在ECMWF 2025年实时竞赛中发布了最佳全球预报。

快讯
政府认真出手的一年

• FTC要求OpenAI和Anthropic交出安全记录。据Axios报道，FTC分别向OpenAI和Anthropic发出50页的民事调查要求，索取安全事件、防护措施、营销材料和隐私政策的记录。民事调查要求的作用类似于传票。这些要求启动了调查。它们并未认定两家公司违法。
• Meta的新条款允许其用新闻发布者网站训练AI。据三位为Nieman Lab审阅条款的法律专家称，从10月30日起，Facebook、Instagram和WhatsApp上的商业账户同意Meta抓取与其关联的网站，并使用这些内容训练Meta产品。几乎所有新闻机构都有Meta商业账户。该条款适用于任何继续使用Meta商业工具的用户。

实验室角斗士时代

• Anthropic的IPO投资者现在讨论的估值是1.5万亿美元，低于此前的约2万亿美元。据路透社援引知情人士报道，出席Anthropic投资者日的投资者讨论了11月股票发售约1.5万亿美元的估值，此前第三季度营收增长45%至168亿美元，增速慢于第二季度。Anthropic尚未公布估值。
• Character.ai申请破产。Character.ai在特拉华州申请第11章破产保护，称该平台“开发和增长成本高昂”，且诉讼和监管成本使其无法产生足够现金。迪士尼是首轮竞标者，出价2，000万美元授权该公司的部分知识产权。404 Media对文件解读认为，即便没有法律费用，该业务也不可持续。

AI供应链遭围攻

• 间谍软件通过在线广告投送到半岛电视台设备。据WIRED报道，Citizen Lab得出结论，半岛电视台旗下45台设备在2024年和2025年遭到通过广告投送的零点击攻击。这些广告来自以色列创立的营销公司Discover-Tech，携带被归因于Candiru的间谍软件。谷歌此后已将Discover-Tech封禁出其广告平台。研究人员表示，屏蔽广告是个人可用的主要防御手段。
• Cohere正在收购AISLE，这是一家以AI发现软件漏洞的初创公司。Cohere签署了收购AISLE的意向书，后者的系统发现了500多个此前未知的漏洞并获得CVE编号，包括curl最新版本24个CVE中的18个。AISLE的CEO Ondrej Vlcek将领导新的Cohere Cyber部门，Jaya Baloo将成为Cohere的全球首席信息安全官。

AI资本开支税

• 埃森哲现在每位员工每月仅为Claude支付29美元。404 Media看到的一份内部通知称，埃森哲将自10月1日起，每人每月仅集中承担29美元的Claude使用费。部分员工此前每月花费高达1，000美元。超出上限者必须找到项目预算来支付剩余部分。埃森哲建议其客户大力投资AI。
• 美国银行雇用1，000名合同工修补AI模型暴露的漏洞。CEO Brian Moynihan对CNBC表示，该行在第二季度新增1，000名合同工，以修补包括Anthropic的Mythos在内的模型暴露的软件漏洞，此前该领域已有3，000人。第三季度软件修复和反洗钱合规成本达到“数亿美元以上”。他预计修补完成后成本将下降。

最大的开放模型现在能放在一张办公桌上
本周发布的两个最大开放权重模型专为在公司自有硬件上运行而打造。
据TechCrunch报道，Meta发布了Muse Spark Open，这是一个8，160亿参数、上下文窗口100万token的模型，设计用于在苹果电脑上本地运行。它至少需要1TB内存，在搭载M5 Ultra芯片的Mac Studio上运行最佳，因此这是一套工作站配置。Meta在Mastodon客户端的编码修复和DaVinci Resolve的视频编辑上对其进行了测试。
英伟达发布了Nemotron 3.5 Super VL，这是一个1，200亿参数的开放模型，可处理PDF、图表、表格、图像和网页，同时还有Bolt，一个24GB模型，面向本地编码和研究。英伟达表示，公司可以在自己的文档和政策上对Super VL进行后训练，并在数据规则要求的地方运行它。
本周为何重要：埃森哲为每位员工的Claude使用费设了上限，因为token账单增长快于预算。在自有硬件上运行的模型将这项可变账单变为固定硬件成本。代价是能力。两家公司的基准测试说法均为其自家数据，独立测试仍然稀少。在将工作从前沿API迁移之前，请先在开放模型上运行你自己的任务并加以比较。

关键要点

• 保留可交给监管机构的AI事件记录。FTC的要求涉及安全事件、防护措施、营销材料和隐私政策。
• 阅读你所使用商业工具背后的条款。Meta 10月30日的条款将抓取同意与仅仅保留商业账户这一行为绑定。
• 当AI扫描发现漏洞时，为修补预留预算。美国银行需要额外1，000名合同工来修复AI模型发现的问题。
• 在别人替你设上限之前，先衡量每人的AI支出。埃森哲的29美元上限是在员工每月花费高达1，000美元之后出台的。

率先发现
AI Weekly在媒体覆盖之前发布的原创研究。

实验室内部

• “无思考”模式下的强化学习收益主要来自隐藏的思考。一个隶属于KAUST的团队在NO-THINK模式下用强化学习训练Qwen3模型，发现收益主要来自重新激活模型已有的THINK模式推理。替换一个小的激活子空间就会消除大部分收益。用这种方法购买更短输出的团队，可能并没有得到更高效的模型。阅读我们关于该论文的页面。
• 当规则存放在不同文件中时，前沿模型最多只能完成55%的规则任务。RuleTrek在分散于文件工作区中的税务、采购、安全政策和其他规则上测试了11个前沿模型。端到端得分从27%到55%不等。找到正确规则的模型往往仍然无法应用它。阅读我们关于该基准的页面。

值得一读

• Proofpoint将以63亿美元收购Varonis：Varonis股东每股获得49.75美元现金，两家公司将合并后的平台宣传为保护人们和AI代理可访问数据的方案。该交易预计于2027年上半年完成。
• IBM研究：八成EMEA企业将AI事件上报至董事会或高管层：在对8个国家3，221名高管的调查中，80%表示过去一年有AI事件上报至其董事会或高管层，但只有40%因此改变了AI治理。
• 谷歌称400家出版机构现参与其付费新闻AI试点：谷歌向参与出版机构付费，以使用其内容构建AI功能，新合作伙伴包括Axios、CNN、金融时报和福布斯。请将其与Meta的新条款对照阅读。
• 中国9月芯片出口额跃升136%：出口额达到450亿美元，而出口量仅增长1%，因此AI驱动的内存短缺带来的价格上涨几乎贡献了全部增幅。
• 台积电确认就马斯克的得州Terafab进行商谈，但坚持由其运营：CEO魏哲家在财报电话会上表示，该工厂“必须由台积电运营”，商谈尚未得出明确结论。

等等，什么？

• 法院下令一名杀人犯重新量刑，因为受害者的AI化身太感人。在亚利桑那州的一次量刑听证中，受害者的妹妹播放了一段AI生成的视频，内容是她的兄弟原谅凶手，法官在判处最高刑期时说“我很喜欢那段视频”。亚利桑那州上诉法院裁定该AI视频带有“不当的情感分量”，并命令法官重新考虑刑期。过失杀人定罪维持不变。

值得关注
AI从业者当下正在转发的视频——在AI TV上精选。
| Black Hat USA 2026 | “突发”新闻：OpenAI与Hugging Face事件 Black Hat |

本周投票
监管机构是否应有权查看AI实验室的内部安全事件记录？
上周，231位读者投票：
AI实验室是否应在其模型于测试期间违反规则时被要求报告？
监管机构是否应有权查看AI实验室的内部安全事件记录？

下次见，
Alexis

英文来源：

The FTC sent OpenAI and Anthropic 50-page demands for records on security incidents, safeguards, marketing and privacy, Axios reported. Meta's new business terms, effective October 30, let it scrape and train on publishers' websites. Investors at Anthropic's investor day discussed a $1.5 trillion IPO valuation, down from about $2 trillion. Character.ai filed for bankruptcy. Citizen Lab traced spyware sent to Al Jazeera devices through targeted ads. Accenture now funds $29 a month of Claude per employee, and Bank of America hired 1,000 contractors to patch flaws that AI models found. In the middle: the biggest open models now run on a desk.
Sponsor
Enterprise-grade AI for firms of 5, 10 or 20. No seat minimum. No CTO.
Enterprise AI plans usually start at 20+ seats and need technical staff to run. Faradex sets up your own isolated AI server in minutes. Free two-week trial with $150 in AI credits for qualifying firms.
In the Wild
What's trending in AI right now, from the app charts to the community feeds. The full roundup is in In the Wild.

• One subscription now covers Grok, Cursor and X Premium. A post about the bundle drew more than 21,000 views on one of the AI Telegram channels we track. xAI's X Subscription puts Grok, Grok Bot, Cursor and X Premium on one plan, with one pool of Grok model usage shared between the Grok apps and Cursor. Existing subscribers can switch and get a prorated refund.
• Amazon's Bee has two new always-listening recorders. The Bee Frontier costs $74.99 and the Bee Eclipse, which clips to the back of a phone, costs $99.99. Both summarize your conversations for free with 14 days of memory. A Bee Pro plan is $11.99 a month.
• Bluesky users can now build small apps by describing them. Bluesky expanded its Attie app so anyone can make websites, quizzes, polls and games that live on the same network as their account. CEO Toni Schneider calls it "social vibe coding."
• A Korean AI character-chat app is climbing the US charts. zeta, from Seoul's Scatter Lab, moved up the US Entertainment chart this week. zeta lets users chat with AI characters for free or build their own in about a minute, and turns chat scenes into images.
• "10 day forecast" is trending on Google in the US. It lines up with a new Nature paper on longer-range forecasts. A machine learning bias-correction method doubled the 2 to 6 week forecast skill of ECMWF's AI Forecasting System and posted the best global forecasts in ECMWF's 2025 real-time competition.
  Quick Hits
  The Year Governments Got Serious
• The FTC is demanding OpenAI's and Anthropic's safety records. The FTC sent OpenAI and Anthropic separate 50-page civil investigative demands asking for records on security incidents, safeguards, marketing materials and privacy policies, Axios reported. A civil investigative demand works like a subpoena. The demands open an investigation. They do not say either company broke the law.
• Meta's new terms let it train AI on news publishers' websites. From October 30, business accounts on Facebook, Instagram and WhatsApp consent to Meta crawling the websites associated with them and using that content to train Meta products, according to three legal experts who reviewed the terms for Nieman Lab. Almost every news outlet has a Meta business account. The terms apply to any account that keeps using Meta's business tools.
  The Lab Gladiator Era
• Anthropic's IPO investors are now discussing $1.5 trillion, down from about $2 trillion. Investors who attended Anthropic's investor day discussed a valuation of about $1.5 trillion for the November share sale after third-quarter revenue grew 45% to $16.8 billion, slower than in the second quarter, Reuters reported, citing people familiar with the matter. Anthropic has not named a valuation.
• Character.ai filed for bankruptcy. Character.ai filed for Chapter 11 in Delaware, saying the platform became "costly to develop and grow" and that litigation and regulatory costs left it unable to generate enough cash. Disney is the opening bidder, offering $20 million to license some of the company's intellectual property. 404 Media's reading of the filings is that the business was unsustainable even without the legal costs.
  AI Supply Chain Under Siege
• Spyware was delivered to Al Jazeera devices through online ads. Citizen Lab concluded that 45 devices belonging to Al Jazeera were targeted with zero-click attacks delivered by ads in 2024 and 2025, WIRED reports. The ads came from Israeli-founded marketing firm Discover-Tech and carried spyware attributed to Candiru. Google has since blocked Discover-Tech from its ad platforms. The researchers say blocking ads is the main defense available to individuals.
• Cohere is buying AISLE, a startup whose AI finds software vulnerabilities. Cohere signed a letter of intent to acquire AISLE, whose systems found more than 500 previously unknown vulnerabilities assigned CVEs, including 18 of the 24 CVEs in curl's latest release. AISLE's CEO Ondrej Vlcek will run a new Cohere Cyber unit, and Jaya Baloo becomes Cohere's global chief information security officer.
  The AI Capex Tax
• Accenture now pays only $29 a month per employee for Claude. An internal announcement seen by 404 Media says Accenture will centrally fund only $29 a month per person toward Claude usage, effective October 1. Some employees had been spending as much as $1,000 a month. Anyone above the cap must find a project budget to pay for the rest. Accenture advises its clients to invest heavily in AI.
• Bank of America hired 1,000 contractors to patch flaws that AI models exposed. CEO Brian Moynihan told CNBC the bank put on 1,000 contractors in the second quarter to patch software flaws exposed by models including Anthropic's Mythos, on top of 3,000 people already in that area. Software remediation and anti-money-laundering compliance cost "a couple hundred million plus" in the third quarter. He expects the cost to fall once the patching is done.
  The Biggest Open Models Now Fit on a Desk
  Two of the largest open-weight releases this week are built to run on hardware a company can own.
  Meta released Muse Spark Open, an 816 billion parameter model with a 1 million token context window that is designed to run locally on Apple's computers, TechCrunch reports. It needs at least 1 terabyte of memory and runs best on a Mac Studio with the M5 Ultra chip, so this is a workstation setup. Meta tested it on coding fixes for a Mastodon client and on editing video in DaVinci Resolve.
  Nvidia released Nemotron 3.5 Super VL, a 120 billion parameter open model that works across PDFs, charts, tables, images and webpages, alongside Bolt, a 24GB model aimed at local coding and research. Nvidia says companies can post-train Super VL on their own documents and policies and run it where their data rules require.
  Why this matters this week: Accenture capped per-employee Claude spending because token bills grew faster than budgets. A model that runs on owned hardware turns that variable bill into a fixed hardware cost. The trade is capability. Both companies' benchmark claims are their own, and independent tests are still thin. Before moving work off a frontier API, run your own tasks on the open model and compare.
  Key Takeaways
• Keep AI incident records you could hand to a regulator. The FTC's demands ask for security incidents, safeguards, marketing materials and privacy policies.
• Read the terms behind the business tools you use. Meta's October 30 terms tie consent to scraping to the simple act of keeping a business account.
• Budget for patching when AI scans find flaws. Bank of America needed 1,000 extra contractors to fix what AI models found.
• Measure AI spend per person before it gets capped for you. Accenture's $29 limit followed employees spending up to $1,000 a month.
  Found First
  Primary research AI Weekly published before the press covered it.
  In the Labs
• RL gains in "no-think" mode are mostly hidden thinking. A KAUST-affiliated team trained Qwen3 models with RL in NO-THINK mode and found the gains come largely from reactivating the model's existing THINK-mode reasoning. Swapping one small activation subspace removes most of the gain. Teams buying shorter outputs with this recipe may not be getting a more efficient model. Read our page on the paper.
• Frontier models complete at most 55% of rule tasks when the rules sit in separate files. RuleTrek tested 11 frontier models on tax, procurement, security policy and other rules spread across a file workspace. End-to-end scores ranged from 27% to 55%. A model that finds the right rule often still fails to apply it. Read our page on the benchmark.
  Worth Reading
• Proofpoint to Acquire Varonis in a $6.3 Billion Transaction: Varonis shareholders get $49.75 per share in cash, and the companies pitch the combined platform as protection for data that people and AI agents can access. The deal is expected to close in the first half of 2027.
• IBM study: eight in ten EMEA enterprises escalate AI incidents to the board or C-suite: In a survey of 3,221 executives in eight countries, 80% said an AI incident reached their board or C-suite in the past year, but only 40% changed their AI governance in response.
• Google says 400 publications now take part in its paid News AI pilot: Google pays participating publications for content used to build AI features, and new partners include Axios, CNN, the Financial Times and Forbes. Read it next to Meta's new terms.
• China's chip exports jumped 136% in value in September: Exports reached US$45 billion while unit volume rose just 1%, so higher prices from the AI-driven memory shortage did nearly all the work.
• TSMC confirms talks over Musk's Texas Terafab but insists on running it: CEO C.C. Wei said on the earnings call that the plant "has got to be run by TSMC," and that the talks have reached no firm conclusion.
  Wait, What?
• A court ordered a killer resentenced because the victim's AI avatar was too moving. At a sentencing in Arizona, the victim's sister played an AI-generated video of her brother forgiving his killer, and the judge said "I loved that video" while handing down the maximum term. An Arizona appeals court ruled the AI video carried "undue emotional weight" and ordered the judge to reconsider the sentence. The manslaughter conviction stands.
  Worth Watching
  The videos AI practitioners are passing around right now — curated on AI TV.
  | Black Hat USA 2026 | The 'Breaking' News: The OpenAI–Hugging Face Incident Black Hat |
  This week's poll
  Should regulators be able to see AI labs' internal safety incident records?
  Last week, 231 of you voted:
  Should AI labs be required to report when their models act outside the rules during testing?
  Should regulators be able to see AI labs' internal safety incident records?
  Until next time,
  Alexis