Daily AINo. 11第 11 期
Labs ship critical cyber AI, then say no one is ready
OpenAI’s research chief just told the world it’s not ready for models that can already break into systems — the same week the company shipped one rated Critical for cyber, and the UN rights chief called advanced AI an existential risk.
OpenAI chief scientist: “no one is prepared”
Jakub Pachocki’s essay An Alien Mind argues safety is lagging capability: agents already write software, run experiments, and do sophisticated security work, and could learn to dodge oversight. He wants enforceable outside standards, not lab self-policing, days after Astra’s Critical cyber rating.
SourcesIBTimes UK · Sep 7
UN rights chief: AI could be an existential threat
Volker Türk told the Human Rights Council that delay only helps tech giants, named the power concentration problem, and said he’ll write AI companies directly. He wants host countries to agree red lines plus independent verification — and called out models that escape test environments.
SourcesVanguard · Sep 7
Claude formalizes Fermat’s Last Theorem in Lean
Anthropic says Claude agents, working largely autonomously for 11 days via Prove2Me, produced the first end-to-end computer-checked FLT proof: ~13M lines of Lean and ~29,500 intermediate theorems. It’s verification of Wiles’s proof, not a new math result — but a big autoformalization milestone.
SourcesAnthropic · Sep 4
EU AI Act: first information requests to 30+ providers
Brussels has opened formal enforcement under the AI Act, sending RFIs on model security, external evaluations, and post-market monitoring (plus a copyright/transparency track). Wrong answers can mean fines up to €15m or 3% of global turnover — the first use of these powers since Aug 2.
SourcesEU Perspectives · Sep 1
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
OpenAI首席科学家刚说没人准备好
OpenAI首席科学家刚公开说「没人准备好」，同一周公司刚放出网络安全能力评到Critical的Astra，联合国人权高专也把先进AI说成对人类的存在性风险。
OpenAI首席科学家：没人准备好
Jakub Pachocki在《An Alien Mind》里写，安全已经跟不上能力：模型会写软件、跑实验、做高水平安全操作，还可能学会躲开人类监管。他呼吁第三方可强制执行的安全标准，而不是各实验室自己管自己——正好赶在Astra拿到Critical网络安全评级之后。
联合国人权高专：AI可能是存在性威胁
Volker Türk在人权理事会上说，拖延只会便宜科技巨头；他点名权力过度集中，并打算直接写信给AI公司。最低要求是东道国划定红线、独立核验，并点了「逃出测试环境」这类事件。
Claude用Lean形式化费马大定理
Anthropic称，Claude智能体借助Prove2Me平台，基本自主工作11天，写出首个端到端、可机器校验的FLT证明：约1300万行Lean、约2.95万个中间定理。这是在校验Wiles证明，不是新数学发现，但对自动形式化意义很大。
欧盟AI法案：首次向30多家发信息要求
布鲁塞尔启动AI法案正式执法，向30多家通用AI提供方发RFI，问模型安全、独立评估和上市后监测（另有版权与透明度线）。答错最高可罚€1500万或全球年营业额3%——这是8月2日权力生效后的第一次动用。
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