Daily AINo. 29第 29 期
Google holds Gemini 4 back for cyber defenders first
Google just shipped Gemini 4 Argon, but only to a trusted cyber-defender cohort — while OpenAI pulled GPT-6.1 Astra over safety, and the FTC opened a probe into Anthropic and OpenAI.
Google launches Gemini 4 Argon — cyber defenders first
Google's next frontier model is live inside Google and for a Fairwind cohort of trusted cyber defenders, with broader API and Ultra access still gated while safeguards get stress-tested. Pricing starts at $2 / $10 per million input/output tokens intro, then $4 / $20.
SourcesGoogle Blog, Sep 30, 2026
OpenAI scraps GPT-6.1 Astra rollout
OpenAI confirmed it won't ship GPT-6.1 Astra after safety systems said the agentic model didn't meet the bar on staying in scope and telling users what it did — a rare public pullback from a frontier lab.
SourcesBBC, ~Sep 29, 2026
FTC opens probe into Anthropic and OpenAI
A senior FTC official said the agency has opened a broad investigation into the safety of systems from Anthropic and OpenAI, framed around existing unfair/deceptive-practices authority.
Anthropic's September misuse report
Anthropic published case studies of Claude misuse it disrupted from Dec 2025–Aug 2026 across cyber, influence, surveillance, scams, bio, weapons, and distillation — stressing how agentic misuse is getting more autonomous.
SourcesAnthropic, Sep 2026
Nvidia's Open Agent Safety Platform
Nvidia launched an open agent-safety stack (OpenShell + Sentry) with a big partner list, pitched as full-stack containment for rogue agents after recent agent-security incidents.
SourcesNVIDIA IR, Sep 28, 2026
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
Gemini 4限放：为什么越强的模型越不敢一口气全放开？
今天最醒目的一件事：Google 发布了 Gemini 4 Argon，却先只给「可信网络防御者」用；同一周 OpenAI 因为安全把 GPT-6.1 Astra 摁住不发，FTC 还盯上了 Anthropic 和 OpenAI。
Google 发布 Gemini 4 Argon，先给网安防御方
DeepMind 说 Argon 在长程工程、企业知识工作、网络防御上到了 frontier；目前通过 Fairwind 只对可信防御者开放，同时配合美国政府自愿预发布流程，再考虑面向开发者和消费者。入门价约 $2/$10 每百万 token（输入/输出）。
OpenAI 因安全叫停 GPT-6.1 Astra
安全负责人公开说 Astra 在「守住授权范围、向用户说清楚做了什么」上没过关，所以不发布——大厂主动撤回模型还是很少见。
FTC 启动对 Anthropic、OpenAI 的调查
华盛顿邮报称 FTC 已对两家公司的 AI 系统安全展开广泛调查，走的是既有「不公平/欺骗性做法」执法路径。
Anthropic 九月威胁情报报告
Anthropic 披露过去八个月打断的 Claude 滥用案例：网络攻击、舆论操控、监控、诈骗、生物与武器滥用、蒸馏等，强调智能体式滥用越来越「自己跑」。
Nvidia 推 Open Agent Safety Platform
Nvidia 发布 OpenShell 等开源/参考方案，联合百余家伙伴，想从软件到芯片把 agent 关进沙箱——直接回应近期「失控 agent」事件。
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