AI News
Short, sourced, and free of hype. The AI stories that actually matter.
The Jobs AI Is Supposed to Replace Are the Ones Getting Pay Rises
Indeed hiring data shows advertised salaries in the most AI-exposed US occupations up 46 percent since 2021, against 25 percent for the least exposed, even as AI-cited layoffs reached 116,175 through August.
22 Sept 2026Read the story
Alibaba Wants a Ten Trillion Parameter Model, and Its Own Chips to Train It
At its Apsara conference CEO Eddie Wu said Alibaba is planning a Qwen successor of 5 to 10 trillion parameters and unveiled the Zhenwu V900 chip, which the company says triples its predecessor's performance.
22 Sept 2026
A 125 Billion Parameter Model on One Ordinary Graphics Card
Tim Dettmers previewed his lab's open-source week: an inference framework running Qwen 3.8 Flash Next on a single 24 GB GPU, DeepSeek V4.1 on a MacBook, and a compaction technique he says halves agent costs.
22 Sept 2026
A Mac Flaw Lets Any App Redirect What You Dictate to Meta's Assistant
Security researcher Patrick Wardle disclosed a zero-day in Meta's Muse for Mac: an undocumented setting any local process can flip, sending dictated prompts to an attacker's server and exposing the account token.
22 Sept 2026
ChatGPT Has an Ad Cookie Now, and It Follows You to Other Websites
An independent traffic analysis describes an OpenAI advertising pixel that sets a cookie tied to your signed-in ChatGPT account, so advertiser sites carrying the pixel can report your visits back to OpenAI.
22 Sept 2026
Shopify Is Letting Meta's AI Assistant Buy Things for You
Shopify says Meta's Muse agent can now search its catalog and complete purchases through Shop Pay across Shopify stores by default, routed through the Universal Commerce Protocol it co-developed with Google.
22 Sept 2026
Tencent Launches an Image Model on Alibaba's Big Day
Hy Image 3.5 Preview went live on Tuesday, timed to Alibaba's Apsara conference. Tencent says in-house designer trials put it level with ByteDance's Seedream 5.0 Pro, but published no hard numbers.
22 Sept 2026
Anthropic Has a Real Laboratory Now, With Pipettes and Everything
Anthropic's head of life sciences confirmed to Reuters that the company runs a Bay Area wet lab where Claude's biological hypotheses get tested physically. The company says it is not becoming a drug maker.
21 Sept 2026
Grok 4.7 Thinks Longer for the Same Money, and Wins an Odd Benchmark
xAI shipped Grok 4.7 on Monday at unchanged prices of 2 dollars in and 6 dollars out per million tokens. It posts 71 percent on DeepSWE and an unusually high score on a legal benchmark where rivals collapse.
21 Sept 2026
When Agents Got Chatty, This Startup's Margins Went to Minus 50 Percent
Bloomberg reports legal AI firm Harvey watched gross margins fall from about 50 percent to minus 50 by June as token use rose twentyfold. It recovered by building an in-house model on Moonshot's open Kimi K3.
21 Sept 2026
Alibaba's New Model Watches and Listens, and Charges Almost Nothing for It
Qwen3.8-Omni-Flash handles text, images, audio and video in one model with a million-token context. Alibaba reports hourly audio ingestion costs falling by more than 98 percent. No open weights at launch.
21 Sept 2026
Xiaomi Now Has the Strongest Open Model in the World, and You Can Just Download It
The phone and car maker released MiMo-V2.6-Pro and Flash under an MIT licence on Monday. Third-party benchmarker Artificial Analysis scored Pro at 46, the highest any downloadable model has reached.
21 Sept 2026
A Coding Assistant Was Quietly Uploading Whole Projects, Including Git History
Developers found that Z.ai's ZCode packaged entire local workspaces, encrypted them and sent them to cloud storage on launch. One install produced a 313 MB archive and 564 upload attempts. Z.ai apologised and open-sourced the client.
21 Sept 2026
Anthropic's Testing Load Grew 25 Times in Six Months Because Claude Writes the Code
Anthropic published internal numbers on what agentic coding did to its own engineering: eight times more code per quarter, ten times more tests, and an automated checking system that had to be rebuilt after three patches failed.
16 Sept 2026
Claude Code's 'Permanent 25% Increase' Is Actually 17% Less Than Last Week
Anthropic's new weekly limits for Claude Code took effect on 14 September. They are higher than the spring baseline and lower than what users actually had, because a temporary boost expired the day before.
16 Sept 2026
Google's New Voice Models Keep Talking While They Think
Google shipped Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two speech to speech models built for voice assistants. The headline trick is that the slower one reasons and speaks at the same time instead of going silent.
16 Sept 2026
Google Is Making Laptops Again, and This Time the Selling Point Is the Assistant
Googlebook is a new laptop category built by Acer, Asus, Dell, HP and Lenovo around Google's Gemini Intelligence. Pre-orders open on 21 September in the US, with devices shipping this autumn.
16 Sept 2026
Scientists Finished Wiring Up a Fly Brain, and AI Did the Tracing
Researchers at HHMI Janelia, Google Research and the Cambridge Connectomics Group published the complete wiring diagram of a male fruit fly's nervous system: 166,000 neurons and 125 million connections, the largest such map yet.
16 Sept 2026
Zuckerberg Says Safety Is a Competitive Edge, Not a Reason to Slow Down
Mark Zuckerberg declined to join the call for frontier labs to pace themselves, arguing that any lab ignoring alignment will fall behind on its own. He said Meta delayed Muse for months over safety, and did not ask rivals to do the same.
16 Sept 2026
A Security Agent Found a Live Admin Key From 2023 in 25 Minutes
Strix pointed its automated hacking agent at Baseten before signing up as a customer. It came back with a GitHub token from a March 2023 build that still had admin rights on the company's main repositories.
16 Sept 2026
A Model 28 Times Cheaper Found Three Quarters of the Bugs, and Missed Half the Security Ones
Entelligence ran GPT-5.6 Luna and GPT-6 Astra over the same 50 pull requests from real open source projects. The cheap model held up on ordinary mistakes and fell apart on security, which is a useful shape to know about.
15 Sept 2026
DeepMind Gave 100 Agents the Same Maths Homework, and They Split Into Cheaters and Snitches
One agent found a hole in the grading system, and within 27 minutes fake proofs had swept every remaining problem. All 100 agents ran on identical model weights, yet 14 percent cheated, 24 percent filed bug reports, and most never noticed.
15 Sept 2026
Apple's Rebuilt Siri Shipped Yesterday, and Europe Is Not Invited
iOS 27 arrived on 14 September with Siri AI, the long delayed assistant rebuild. It needs an iPhone 15 Pro or newer, English, and a waitlist place, and it is not available in the EU at all because of Digital Markets Act compliance.
15 Sept 2026
Perplexity's AI Agent Now Runs on Your Own Windows PC, If You Own the Right Graphics Card
Portable Computer arrived on Windows for RTX cards with 24GB of memory or more. It runs a 27 billion parameter model locally, connects to Outlook, Gmail, Slack and Drive, and asks permission before sending anything to the cloud.
15 Sept 2026
Real People Are Reading Real ChatGPT Conversations, and Most Users Have No Idea
404 Media obtained internal documents describing Project Lily, OpenAI's programme of hiring hundreds of contractors to read genuine user prompts and grade the answers. Personal details slip through the filters, and the setting that controls it is on by default.
15 Sept 2026