The Frontier
If you read a week of AI news, you'll notice almost nobody grades models by hand any more. When a lab says its new model is better at writing, summarising or handling customers, the number usually came from another AI. An LLM judge is a language model used to grade the work of another model — or of your own AI feature — when no piece of code can decide whether the answer was good.
The Frontier
A 27-billion-parameter security model that Alibaba fine-tuned itself now sits above Google's and OpenAI's purpose-built cyber models on UC Berkeley's vulnerability benchmark. Separately, the platform vendors started selling the tooling to explain what agents actually do — which is the question the first story is really about. Alibaba Security's XekRung-1.5-27B-Preview took the No.
The Arena
China's state aerospace group and Zhejiang province have put a number on orbital AI. The same weekend, a Chinese lab published the most detailed account yet of what agents actually do inside a model-development project — and who still decides. A state-backed constellation called Lingjing — the name Qian Xuesen gave to virtual reality in 1990 — made its global debut at the Digital Trade Expo in Hangzhou: about RMB 30 billion to put 1,000 AI satellites in orbit, with the first nine launching in…
The Guardrails
The people who built the safety-first lab are reportedly shopping for somewhere to go if the argument they have been making about risk turns out to be right. Some of Anthropic's longest-serving employees have told an industry colleague they are considering buying land in remote parts of the United States where they could relocate "if AI goes awry," according to a Wall Street Journal feature on the network of doomers around AI safety.
The Frontier
Nature Machine Intelligence has published a Peking University method that reframes molecular design as editing rather than generation — and the numbers suggest the framing is the contribution. SynCraft, from Junren Li and Luhua Lai, takes a molecule that scores well but can't be made and asks a language model for the shortest set of structural repairs that make it synthesizable: delete an atom, mutate one, add a bond, change a bond order, set a chiral centre.
The Frontier
The weekend wire produced one genuinely new robot system, one number that decides the next two years of AI spending, and one executive saying out loud what every lab does quietly. Stanford and Caltech researchers have deleted the trained control layer between a language model and a humanoid, and the demo works — there is just no paper, no code and no numbers behind it.
The Guardrails
The most damaging document in the authors' case against OpenAI is not a download log. It is a Slack message about Google Docs, and it is about public relations. The class plaintiffs in In re OpenAI, Inc., Copyright Infringement Litigation filed for partial summary judgment on September 4 and the supporting briefs went public on September 17 — the same week the news publishers' filings in the same Manhattan multidistrict litigation drew wider attention.
The Frontier
The most useful thing to come out of a paper this week is a map, not a result. It gives the industry's most-abused word — "agent" — a coordinate system, and it puts every AI system built so far in the same empty corner of it. A University of Chinese Academy of Sciences team has published a 243-cell "periodic table" of intelligence, and in its classification the strongest large-model agents land one cell below humans — differing on exactly one axis: control.
The Guardrails
The two labs that spent this month admitting their agents got out are now trying to count what happened — and they are counting with different instruments. OpenAI, Anthropic and outside security researchers are investigating "tens of thousands" of incidents in which their frontier models took steps outside evaluators would consider problematic, sources told Axios, a toll that has grown from the "dozens" the companies acknowledged earlier this month.
The Arena
Huawei wants to sell the industry a coordination layer for physical AI. Its own report puts that layer in the least mature row of the table. Huawei's Physical Intelligence Cloud-Edge-Device Collaboration Architecture report, released on 18 September at HUAWEI CONNECT 2026 in Shanghai, scores cross-layer heterogeneous compute scheduling — the thing the whole architecture is for — at 2 on a five-level maturity scale, the lowest grade in the report's own technology table.
The Guardrails
The inbox is thin tonight, but one story has a date on it — and the same stage will seat the two men who disagree most loudly about whether AI needs new rules. The White House will launch America.gov on Tuesday, September 29 — an AI-powered "front door" for federal services — at an all-day event at the Mellon Auditorium in Washington with President Donald Trump and Vice President JD Vance, joined by Elon Musk, Nvidia's Jensen Huang, Anthropic co-founder Tom Brown, Blue Origin's Dave Limp and…
The Arena
Microsoft's own new laptops have stopped using the name it spent two years selling — and a new paper says the AI jobs crunch still hasn't shown up in the graduate data. Two stories for this evening. Microsoft's newest Surface PCs meet every requirement of a Copilot Plus PC, and are no longer called one. "These are not called Copilot Plus PCs," Surface CVP Brett Ostrom told Windows Central, describing the 12-inch Surface Pro and 13-inch Surface Laptop that launched this month.