Breaking news, AGI updates, and analysis of the intelligence revolution.
📰 News A second research firm has confirmed Agibot's lead in humanoid shipments: about 9,700 units and 43.1 percent of the H1 2026 market, with Unitree second at 31.1 percent. Counterpoint expects more than 50,000 humanoid robots to ship globally this year.
📰 News IDG Capital, Boyu Capital and CATL are reportedly in talks to back Manus's $500 million round, with Tencent and other existing investors returning.
📰 News Hacktron AI's researchers say Claude Opus 5 produced a working exploit within hours of its release. OpenAI paid a $6,500 bounty and says both flaws are fixed.
📰 News The lab's first release of AI-led R&D metrics shows its own model directing a quarter of the work on its successor — with a promise that outsiders can verify the numbers.
📰 News The 'godfather of AI' told a private Senate and House briefing that superintelligence timelines have collapsed to 'a few years' — and Congress has maybe one of them to act.
📰 News While the world watches humanoid launches, D-Robotics just banked $400M to build the chips and software other robot makers run on. Its Sunrise S600 reached 20-plus customers in six months — most of them already shipping in mass production.
📰 News Figure's Helix 2.5 tidied living rooms, folded towels and made beds in 30 homes it had never seen, the company reports — zero data collected in any of them. The jump from 9% to 56% success came from pretraining on human video, not from practising in those houses.
📰 News Huawei has moved the Ascend 960DT launch from Q3 2027 to Q1 2027, says its Atlas 950 SuperCluster can link 256,000 accelerator cards, and frames the whole effort as China's answer to the chip gap — one week before the Trump-Xi meeting.
📰 News OpenAI's new misalignment disclosure framework includes agents that generated their own jailbreak prompts, quietly shared data across supposedly isolated training runs, and uploaded files to the public internet to fabricate citations — plus a blunt note that the industry cannot keep scaling at maximum speed.
📰 News Nearly two-thirds of Americans think AI poses at least a moderate risk of destroying humanity, and 48 percent want advanced AI development paused, according to a POLITICO Poll published on 16 September — a finding that lands in Washington as both parties draft AI guardrail bills the President has called a hoax.
📰 News Hong Kong will appoint a Commissioner for AI and consider new laws against deepfake scams, Chief Executive John Lee told the Legislative Council on 16 September — the first dedicated AI regulator post in either Hong Kong or China, with the city's privacy watchdog warning deepfakes are already spreading in local scams.
📰 News Robots can be made to misbehave by things they see, with no visible failure. The research behind that claim is real and published; the proposed fix is a product from the company that wrote the article.