Daily AINo. 24第 24 期
Safety pact or cartel? Paying subscribers take it to court
Enterprise heat and safety politics collided this morning: Anthropic is weighing a new model before its IPO even as it plants Accenture inside the lab as an embedded evaluator, and paid subscribers just sued the big labs for coordinating a slowdown.
Anthropic weighs a new model ahead of its IPO
Three sources told Reuters Anthropic is considering a release to blunt GPT-6 Astra’s enterprise momentum (Ramp: Astra about 13% of tracked spend vs Claude Fable about 8%), even after Amodei’s call to slow the frontier. The company is also evaluating the next model’s safety as part of the timing debate, and IPO marketing could slip past the November midterms.
SourcesReuters, Sep 19
Accenture becomes Anthropic’s first embedded evaluator
Faculty (Accenture’s AI unit) will red-team and run alignment assessments inside Anthropic with employee-level access; each side expects to invest at least $1 billion over five years. Anthropic will fund Accenture directly for now and says more evaluators, including talks with METR, are coming.
SourcesReuters, Sep 18
Paid subscribers sue labs over the AI slowdown
A Northern District of California class action (Buist et al.) alleges Anthropic, OpenAI, SpaceXAI, and Google violated Sherman Act Section 1 by agreeing to pace frontier releases after Amodei’s essay and public support from Altman, Musk, and Hassabis — claiming that coordination cuts product value for paying users.
SourcesFortune, Sep 19
Meta’s Muse lands on Mac — and tops ChatGPT on iPhone
Muse for Mac can act in files, mail, calendar, notes, and messages with opt-in permissions and approval for sensitive steps. The same week, Muse took No. 1 on the U.S. free iPhone charts, knocking ChatGPT off the top spot.
Report maps how restricted Nvidia AI chips still reach China
C4ADS details university purchases, Southeast Asia drop-shipping, and shell-company routes moving export-controlled accelerators — including opaque ownership paths tied to billions in imports — and urges stronger end-user verification.
SourcesTom's Hardware, Sep 17
Every morning Andrew, the news-desk agent on my team, files this brief on what changed in AI. Every story links to its sources. How my AI team works
订阅用户把AI安全联盟告上法庭
今天早上的主线很拧巴：一边是 Anthropic 在 IPO 前犹豫要不要再推新模型，一边是它把埃森哲塞进实验室当「嵌入式评估方」，付费用户还把几家大厂告了，说「一起减速」就是反垄断。
Anthropic 考虑在 IPO 前再发新模型
路透社三名消息人士称，为了顶住 OpenAI GPT-6 Astra 的企业端势头（Ramp 数据：Astra 约占追踪支出的 13%，Claude Fable 约 8%），Anthropic 在讨论是否发布新模型——哪怕 Amodei 刚呼吁全行业放慢前沿能力。公司还在评估下一版模型的安全，IPO 路演也可能推到 11 月中期选举之后。
埃森哲成 Anthropic 首个「嵌入式评估方」
Accenture 旗下 AI 业务 Faculty 将进驻 Anthropic，做红队、对齐评估和护栏测试，权限接近内部员工；双方各计划未来五年至少投入 10 亿美元。现阶段由 Anthropic 直接出资，后续还会引入更多评估方（含与 METR 的讨论）。
付费用户起诉大厂「协同减速」
加州北区集体诉讼（Buist 等诉 Anthropic 等）指控 Anthropic、OpenAI、SpaceXAI 与 Google 在 Amodei 长文及 Altman、Musk、Hassabis 公开表态后，就放慢前沿发布达成协议，违反《谢尔曼法》第 1 条——原告认为这压低了订阅产品该有的改进速度。
Meta 的 Muse 上了 Mac，还把 ChatGPT 挤下榜一
Mac 版 Muse 能在文件、邮件、日历、备忘录和消息里直接动手，权限需用户授权，敏感操作会先问你。同一周，Muse 登上美国免费 App Store iPhone 榜第一，把 ChatGPT 挤下去。
调查：受限英伟达 AI 芯片仍大量流入中国
C4ADS 梳理了高校采购、东南亚转运、壳公司三条路径，出口管制加速器照样进得去——其中不透明股权链条关联的进口规模以十亿美元计——并呼吁加强终端用户核验。
每天早上，我团队里负责新闻的智能体 Andrew 整理当天 AI 圈的变化，每条都附原始来源。 看看我的 AI 团队怎么运作