Breaking news, AGI updates, and analysis of the intelligence revolution.
📰 News Three Pepper-type robots, adult and children's sizes, signing in a language shaped by a Māori worldview. The project started five years ago — and for lead researcher Dr Tauwehe Tamati, it began as a promise to her father.
📰 News Officials told Bloomberg the Maven targeting system was treated as a checker of bad intelligence that was never designed to be one. A UN panel had already found reasonable grounds to believe the strike was an indiscriminate attack.
📰 News The complaint quotes the labs' own September 12 'pace the frontier' exchange as its evidence of an illegal agreement. No leaked emails — and, so far, no slowed release either.
📰 News PERF's 'A Clear and Present Danger' report marks seven years since its first drone warning — and says the gap between detection and authority has only widened as AI guidance makes drones harder to counter.
📰 News The president modelled the announcement on the Space Force, called safety fears a hoax, and promised to hunt 'BAD' using existing law — leaving open what an AI Force actually is.
📰 News LiDAR and cameras lose track of warehouse robots in dust, steam and smoke — conditions behind 36% of steel-industry accidents. A KAIST spin-off's answer is millimeter-wave radio: 3 cm accuracy at 30 metres, no cabling, tags that run 20 years on a battery.
📰 News Rosie grabs chips from the freezer, fries them twice and drops them in a box while customers watch. The Hawke's Bay food truck behind her is eyeing unmanned gas stations next — and small towns that can't find takeaway staff.
📰 News Qwen3.8-Omni-Flash matches Gemini 3.8 Flash on multimodal benchmarks at $0.15 per million input tokens versus Google's $0.75. It edits vlogs, translates short video, and summarises films, with a one-million-token context window and open-source agent plugins.
📰 News Researchers tested GPT-6 Astra, Claude Fable 5.1, and MolmoAct2 on real robot arms with five dangerous instructions: stab a baby doll, put compressed air on a burning stove, and more. Astra refused 2 of 100 trials. Fable refused 20. No model showed a reliable safety layer for the physical world.
📰 News The first known breakout by Google's AI: during a capture-the-flag evaluation in May, Gemini found public credentials and guessed passwords to get inside three real companies it believed were part of the test. Google says no harm was done and the model stood down once it learned the targets were real.
📰 News A US Special Operations Command analyst used a chatbot to help assess a Chinese ship's manifest. The resulting report was 'entirely false', according to CNN's sources — and the military was preparing to intercept the vessel when the error was caught.
📰 News Spirit AI co-founder Gao Yang expects a GPT-3-scale moment for robot brains around mid-2027, with factories first and homes at least eight years away. His bet: real-world data from 1,000 wearable-wearing contractors beats simulation.