The AI Weekly · Week of November 9, 2026
Frontier labs hoard the real frontier as capital, benchmarks and trust all wobble
- Anthropic's Unreleased Models Outpace Opus 5.5 as External Verification FadesA\三大内部模型曝光！据报道领先Opus5.5三个月，RSI缩短80%实验时间 Dario刚喊减速… 听雨 发自 凹非寺 量子位 | 公众号QbitAI Anthropic的三个内部神秘模型，曝光了。 据METR报告披露，这三个未发布的模型在7月至10月间完成了训练。 其中最强的Model 4，据报道在Anthropic的综合能力指数AECI上，已经超过了Opus5.5。 Anthropic联合创始人Sam McCandlish也出面回…
- Recursive Self-Improvement Is Real But Bounded — and Everyone Is Pricing the Unbounded VersionJeff Dean had just described about 20 automated improvement loops already running at his new company, Discovery Loop, when Fei-Fei Li asked him about the physical limits of recursive self-improvement. His answer was almo…
- Benchmarks Break Both Ways: Zeros, Walls, and a Caught Model CardClaude Opus 5.5 is reportedly the best all-around model on GDPx, and it scores 12%. Every model tested scores zero in finance and legal under binary grading. In the same week, GPT-6 Astra reportedly proved 70% of researc…
- AI Agents Outgrow the Code Supply Chain They Depend OnArticle — "AI Agents Outgrow the Code Supply Chain They Depend On": On November 3, GitHub's status page logged two incidents in a single day. One of them affected Git operations — the plumbing that every clone, fetch, pu…
- AGENTS.md Turns Plain Text Into the New Attack SurfaceHere is the exact moment of no return. You open a webpage, click nothing, download nothing — and a plain-text file called AGENTS.md is already sitting in your Downloads folder. Then you do the most ordinary thing in the …
Threads we are following
- Anthropic's promised progress analysis, including its position relative to its Responsible Scaling Policy and any new safety measures, is due in the coming weeks.
- Whether Anthropic's next model actually crosses its Responsible Scaling Policy threshold for 'dramatically accelerated progress attributable to AI,' and whether METR is allowed to publish the evidence.
- Whether the next round of benchmarks can distinguish refusal-driven zeros from capability-driven zeros without relying on lab self-reporting.
- Whether GitHub's staged write-queue and read/write separation changes will restore merge reliability before more labs migrate their primary repositories off it.
- Whether OS and agent vendors will introduce provenance or origin controls for instruction files, or continue treating folder contents as implicitly trusted.
One issue a week. Free.
sift. reads hundreds of sources so you don't have to, and sends one issue you can actually finish. No daily noise. Unsubscribe anytime.