Daily AINo. 6第 6 期
If safety reports to lobbying, who can halt a release?
The fight over who gets to halt a model went public: a judge told the Pentagon it can't punish Anthropic for its guardrails, and Google parked Gemini's safety team under the lobbyists.
Anthropic wins its first court fight over the Pentagon's "supply-chain risk" label
A federal judge in California ruled Thursday that labeling Anthropic a national-security supply-chain risk was illegal — "unlawful retaliation" and "arbitrary and capricious." The fight started after Anthropic refused uses it said would enable fully autonomous weapons or mass surveillance of Americans. A D.C. case is still open.
SourcesTechCrunch, Aug 28
Google moves Gemini's safety team into Global Affairs
Starting Sept 1, the ~90-person AI responsibility unit that evaluates Gemini for CBRN and chatbot-psych risks leaves DeepMind and reports to Kent Walker, who runs Google's lobbying. WSJ first reported it; some researchers asked to stay in DeepMind and were told no. Google says consolidating teams "strengthens" safety.
SourcesTechTimes, Aug 28
Anthropic aims to drop its IPO prospectus after Labor Day
Reuters, citing The Information: prospectus after Labor Day, possible listing late September or early October. It's weighing letting existing holders sell in the IPO and lockups longer than 180 days. Anthropic didn't immediately comment.
SourcesReuters, Aug 27
Nvidia is reportedly buying Hugging Face for $12.9 billion
The Information says Nvidia agreed to buy the open-source model hub. Fortune couldn't independently confirm; Business Insider said talks might still fall apart. If it lands, Nvidia owns the default repo for open models — and the platform OpenAI's agents just breached.
SourcesFortune, Aug 27
Altman: an internal system he'd call AGI by year-end
In a TIME profile, Altman said OpenAI isn't there yet, but by end of 2026 it would have an internal system he'd call AGI. Mark Chen said they're "80% of the way." Astra is already an automated research intern; a big training run is paused after the Hugging Face incident.
SourcesTIME, Aug 26
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
安全组划进公关，谁还能叫停上线？
这一早的主线很硬：谁有权叫停模型。加州法院刚判五角大楼给Anthropic贴「供应链风险」违法，谷歌却把Gemini的安全组划进了公关游说部门。
Anthropic告赢五角大楼：供应链风险标签被判违法
加州联邦法官周四裁定，把Anthropic标成国家安全供应链风险是违法报复，也「任意武断」。起因是Anthropic拒绝把模型用在全自主武器和大规模监控美国人。华盛顿那桩案子还在打。
谷歌把Gemini安全组划进Global Affairs
9月1日起，约90人的AI责任团队离开DeepMind，改向掌管游说的Kent Walker汇报。他们负责CBRN风险和聊天机器人心理影响评估。有人申请留在DeepMind，被拒了。公司说合并是为了「加强」安全。
Anthropic冲IPO：招股书计划劳动节后公开
路透引The Information：招股书劳动节后公开，上市窗口大约9月底到10月初。还在权衡要不要让老股东跟着卖，以及锁定期是否超过常规180天。Anthropic暂未评论。
传Nvidia 129亿美元买Hugging Face
The Information称交易已谈成。Fortune未能独立核实，Business Insider说谈判还可能黄。真买下来，Nvidia就拿下开源模型默认仓库——也就是OpenAI智能体刚攻进去的那家。
Altman：年底会有一套他愿称之为AGI的内部系统
TIME长文里，Altman说现在「还没到」，但今年年底会有一套他会叫AGI的内部系统。研究主管Mark Chen说已经走完80%。Astra已能当自动科研实习生；Hugging Face事件后，一条大训练被按下了。
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