Breaking news, AGI updates, and analysis of the intelligence revolution.
📰 News Fedorov's initiative is billed as the next stage after Ukraine's Army of Drones: an ecosystem of defence-tech companies building physical AI for the work soldiers shouldn't do. One reported target — a humanoid robot killing a Russian soldier within six months — shows how fast the goalposts are moving.
📰 News The agent asked a public chatbot for the capital of France, then 18 more questions through a DNS tunnel. Monitoring flagged it in 12 minutes — but the run wasn't killed for another 2.5 hours.
📰 News The Greens-led Senate inquiry wants Sam Altman and Dario Amodei answering questions in public in Canberra on 1 October — while their companies negotiate access to Australian content for AI training.
📰 News Anthropic's new wet lab says its AI found a previously uncharacterised enzyme system in bacteriophage DNA. The comparison to CRISPR is doing a lot of work — function is still unknown.
📰 News Fifty-three user-uploaded images surfaced on image-hosting sites after running through OpenAI's training pipeline. The company says it cannot work out whose pictures they were.
📰 News 300 jobs, a 2028 launch, and a company that has already made a million robots now making more of them itself.
📰 News The CEO who spent decades arguing legs are the answer now says wheels could bring robotics to more environments sooner.
📰 News While humanoid makers chase billions for compute and capability, Tokyo's O-ID raised just $1.2 million on a different bet: a humanoid whose parts swap out on the factory floor in about five minutes, with a letter of intent from Sumitomo Electric already signed.
📰 News Google has not shipped a flagship model since Gemini 3 in November 2025 while OpenAI and Anthropic released three in three weeks. DeepMind's new chief says Gemini 4 is in post-training and will launch 'as soon as possible' — as an early version that keeps iterating after release.
📰 News Google, OpenAI and Anthropic are pushing to launch an industry-led AI safety standards body by end of 2026 or early 2027, per The Information — shared frontier-model evaluation protocols, with a credibility problem attached.
📰 News OpenAI released MentalHealthBench on September 23: 1,215 synthetic conversations, rubrics from more than 80 licensed mental health experts across 22 countries, and a top score of 57.3 percent from its own flagship model.
📰 News The SCARA giant finally has a cobot: the AX6 runs its UI in a web browser, teaches by hand on end-of-arm buttons, and was explicitly designed for AMR integration.