September 9, 2026
AI News
The AI news worth knowing this week.
GPT-6 Astra: A new generation of intelligence
OpenAI released GPT-6 Astra on September 3 and 4. The headline feature is computer use: it fills out forms, browses, does research and builds websites on its own, and OpenAI says it finishes those tasks in about half the time of GPT-5.6 Sol. It is rolling out to ChatGPT Plus, Pro, Business and Enterprise, plus the API, Azure and AWS Bedrock.
Read the original →Should promotion depend on how workers use AI?
More companies are tying bonuses and promotions to how much staff use AI. Accenture says promotion now requires it, Disney, Meta, JP Morgan and KPMG run internal AI usage leaderboards, and Coinbase has fired engineers who skipped AI training. Workers worry the extra output just becomes the new baseline.
Read the original →OpenAI claims to have solved maths problem that stumped humans for decades
OpenAI says about 10,000 of its AI agents, running for 88 hours, produced a proof for the Navier-Stokes problem, a fluid-dynamics puzzle open for decades. One catch: the proof covers a variant with an external force, not the exact version the $1 million Millennium Prize asks for.
Read the original →Why this month's Microsoft patch release is a doozy
Microsoft's September update fixes a record 972 vulnerabilities, 112 of them rated critical. Ars Technica reads it as vendors rushing patches out ahead of an expected wave of AI-assisted attacks, now that AI models are finding bugs far faster than before.
Read the original →Hackers are stealing Claude tokens from subscribers
Hackers are using common infostealer malware to grab Claude login sessions from subscribers' computers, then burning through their usage allowance. Anthropic signed affected users out, invalidated the sessions, issued refunds and sent warning emails.
Read the original →Introducing CUDA Rust: Two Tracks for Writing GPU Kernels
NVIDIA now lets developers write GPU code in Rust. Two open-source projects do it: cuda-oxide for the classic thread-based style and cutile-rs for the newer tile-based style. The point is catching memory bugs when the code compiles instead of when it crashes on the GPU.
Read the original →OpenAI researcher allegedly pressured mathematician to drop Anthropic co-author from math breakthrough paper
NYU mathematician Tristan Buckmaster says an OpenAI researcher pressured him to drop his co-author from their Navier-Stokes paper because the co-author works at Anthropic. OpenAI denies seeing their work early and says its own AI-generated proof solves a different variant of the problem.
Read the original →How GPT-5.6 Sol helps run quantum computing experiments
An MIT graduate student hooked GPT-5.6 Sol up to Codex and let it run measurements on an uncalibrated six-qubit chip, analyze the results and tune the qubits on its own. The idea is to hand the repetitive parts of lab work to the model.
Read the original →Dubbing & Open source
Open-source voice and dubbing tools worth a look.
stronghamjji/PersoDub
A desktop app that dubs a video into another language in the original speaker's own voice. It keeps every line inside the time it was originally spoken, without stretching or squeezing the audio to make it fit. There's also a built-in dub agent: tell it "retranslate line 4 so it fits" and it retranslates just that line and regenerates its voice. Everything runs on your own machine.
Read the original →
Edge0-AI/Audio8_TTS
A small text-to-speech model (0.6 billion parameters) that speaks 11 languages and can clone a voice from a short sample. Despite its size, it scores higher than much bigger models on standard tests. Apache 2.0 licensed, with ready-made ONNX and SGLang builds for running it yourself.
Read the original →
truespar/paddock
Truespar/paddock is a Rust-based inference server that supports open models on NVIDIA GPUs with OpenAI- and Anthropic-compatible APIs. It utilizes various precision formats like FP8 and Q4 for efficient inference.
Read the original →This week on X
What AI Twitter was talking about this week.
OpenAI (@OpenAI)
This is GPT-6 Astra. Anything you can do on a computer, Astra can do for you. Fast.
Read the original →Sam Altman (@sama)
GPT-6 Astra is here. We hope it will begin to enable a new generation of entrepreneurship, scientific discovery, and building. We believe it is the best model in the world for computer use, professional work, science, coding, cybersecurity, and more.
Read the original →Google (@Google)
Introducing Gemini 3.8, our best reasoning & coding model yet. By leveraging long-running agentic loops, we’re building on the momentum of 3.7 Flash from just three weeks ago to release two new 3.8 variants: Meet Gemini 3.8 Flash and Gemini 3.8 Flash Cyber
Read the original →Mark Zuckerberg (@finkd)
Muse Spark 1.3 is rolling out today with frontier performance almost too cheap to meter. This is the biggest jump we've made so far on coding and agentic work. Try it in Muse Code and our API.
Read the original →Get the next issue by email
In any of 8 languages. Unsubscribe anytime.