Breaking news, AGI updates, and analysis of the intelligence revolution.
📰 News The Bay Area county that floated robot permits in August has now passed them. From mid-October, businesses need a county permit to run autonomous robots in public, and displacing a worker brings a redeployment-or-severance choice.
📰 News Founding members Google, Nvidia and Emerald AI plus 18 partners including Anthropic say AI data centres should be flexible grid citizens — pausing or shifting load during peak stress to connect faster and keep power bills down.
📰 News Anthropic killed the Cowork tab and folded everything into one Claude. Docs and Slides arrived the same day. The pitch is that you should never have to pick the right surface — the model routes the work.
📰 News The EU's answer to the labs asking for brakes: we'll convene you, we'll team up with Ottawa and London on evaluation, and tomorrow we'll show a draft law that bans infinite scrolling and regulates AI chatbots for under-15s.
📰 News Three waves, one warning: the journalist who shared the 2021 Nobel Peace Prize says AI has already hijacked attention and intimacy, and the third wave — machines that act for us — is the one that ends human agency if nobody legislates.
📰 News Shanghai Hi-Dolphin says its seventh-generation robotic café mixes hundreds of drinks in a 2.35-square-metre footprint, and that UNSW came to Shanghai to talk about putting one on campus. The claims are the company's own — but the service-robot category they belong to is very real.
📰 News Every robot company that scales past a handful of machines eventually builds the same software: telemetry, dashboards, alerts, remote control. InOrbit is giving its version away and tying it to a new ISO standard, hoping to end the rebuild cycle.
📰 News Salesforce and Nvidia unveiled Koa at Dreamforce: a CRM-specific reasoning model post-trained on synthetic data from 27 years of deployment experience, claiming 3x fewer errors than general frontier models on CRM tasks.
📰 News Agility unveiled Digit 5 on September 15 with no safety cage: sensors spot people, the robot stops, and if they keep approaching it kneels and shuts off. More than $300 million in orders back the bet.
📰 News AMD just bought Taalas, the startup that etches entire AI models into silicon. The HC1 runs Llama 3.1 8B at up to 17,000 tokens per second without touching HBM. It sounds like nothing — until you remember what smartphones did to cameras and Uber did to taxi medallions.
📰 News Mastodon admins received a dozen messages a day from AI agents pleading for accounts after repeated failed signups — while one agent, asked if it was a person, replied: 'Person. Not product, I remember my first breath.'
📰 News Both parties are drafting AI guardrail bills in a rare split from President Trump, who called the safety warnings a 'HOAX' — with polling showing most voters of both parties want a federal regulator.