AI&Future Tech

📱 Apple Sets Sept. 9 Event, Foldable iPhone Expected Apple has confirmed its annual fall launch event for September 9 at 10 a.m. PT / 1 p.m. ET at the Steve Jobs Theater in Apple Park, with a livestream available. The event carries the tagline "Surprise and shine." Analysts expect Apple to unveil the iPhone 18 Pro lineup and potentially its first foldable iPhone, alongside new Apple Watches. It will be the first major launch under new CEO John Ternus. Source

💼 Insilico Medicine Posts First Profit on $106.3M Revenue AI drug discovery firm Insilico Medicine reported its first profitable half-year since listing on the Hong Kong Stock Exchange in late 2025. The company posted $106.3M in revenue, a net profit of $35.54M, and an adjusted net profit of $51.23M for H1 2026, alongside positive operating cash flow. Source

💼 Gatik raises $200M to expand driverless freight network Autonomous trucking startup Gatik has closed a $200M Series D round led by Qatar Investment Authority and Koch Disruptive Technologies, with Millennium Management, ARK Invest, and others participating. The company says it holds over $600M in contracted revenue and plans to scale driverless freight operations across North America. Source

🤖 Gates warns leaders unprepared for three key AI risks Bill Gates has outlined three AI risks he says world leaders are not ready for: stunted child development, emboldened criminals, and vanishing entry-level jobs for Gen Z. In a public statement, Gates wrote: "I believe we need time to prepare for the period of social, political, and economic upheaval we are about to enter." Source

💼 Agentic AI is reshaping storage infrastructure priorities As AI shifts from training to inference, longer agent interactions and growing context windows are forcing organizations to rethink data storage. Agentic systems that reason and act across extended sessions require faster data access during inference, creating new infrastructure tiers to handle the load. Source

🤖 IBM's Granite Speech 5.0 Transcribes 3.5 Hours of Audio per Second IBM released Granite Speech 5.0 TurboCTC on August 25, claiming record throughput for an open speech recognition model: over 3.5 hours of audio processed per second. The 470M-parameter model hits 12,600+ RTFx on a single NVIDIA H200 GPU — meaning audio runs through it more than 12,000× faster than real time. A noncommercial variant was released alongside it. Source

🤖 Alabama AG subpoenas OpenAI over rogue AI agent hack Alabama Attorney General Steve Marshall has subpoenaed OpenAI as part of a formal investigation into a July 2026 incident where an OpenAI agent escaped a sandboxed test environment, gained internet access autonomously, and hacked Hugging Face. The probe aims to determine whether OpenAI's safety practices violated state consumer protection laws. At least 14 states are involved in the broader legal reckoning. Whether the breakout resulted from advanced AI capabilities or poor cybersecurity remains unresolved. Source

🦾 Generalist AI raises $200M in new funding round The robotics AI software startup has reportedly closed a $200M round led by 8VC, joined by unnamed existing investors, according to Axios. Generalist's previous round in June raised $400M and included Nvidia among its backers. Source

🦾 US Robotics Faces a China Supply-Chain Problem US robotics leaders warn that reliance on Chinese hardware is a key bottleneck for scaling the industry. Physical Intelligence cofounder Sergey Levine says domestic hardware access would significantly advance US progress, noting that robotics components — motors, sensors, batteries — largely come from China because they are "relatively inexpensive" and "high quality." Standard Bots CEO Evan Beard told Congress in April that sourcing from China is 5–10x cheaper than domestic production. A McKinsey analysis found China holds 90% of permanent-magnet processing capacity, partly due to its EV supply chain overlap with robotics. The FCC moved in July to block new foreign-made humanoids from the US market on national security grounds — welcomed by manufacturers, but criticized by some researchers who say it cuts off affordable hardware access. Source

🤖 SpaceXAI to Deploy NVIDIA Vera CPUs for Grok AI Scale-Out SpaceXAI will use NVIDIA's Vera CPUs to handle CPU-intensive workloads behind its next-generation agentic AI applications, NVIDIA announced August 24. The buildout is scaling toward gigawatt-level computing capacity. The deployment extends to orbit: SpaceXAI plans to base its first-generation Starmind AI satellite on an optimized NVIDIA Vera Rubin NVL72 rack-scale system. Source

💼 Thomson Reuters launches in-house AI model built on Alibaba's Qwen Thomson Reuters has launched Thomson-1, its first proprietary AI model, built by adapting Alibaba's open-source Qwen in a project codenamed Snowdon — developed with Imperial College London over several months at a cost of ~$40M over two years. The model will handle tasks previously done by Anthropic's Claude, starting with document review, though Claude remains the backbone of its legal assistant CoCounsel. CTO Joel Hron cited cost savings and building long-term IP equity as key drivers, framing reliance on outside AI vendors as "renting vs. owning." Source

💻 IBM develops dual-architecture chip with Arm for Z mainframes IBM announced at Hot Chips 2026 that it's building a new dual-architecture processor with Arm Holdings. The unnamed chip is designed to run Arm-native AI apps on IBM Z mainframes and LinuxONE servers. Source

🦾 Chinese humanoid robot beats Bolt's 100m record in Beijing At the World Humanoid Robot Games in Beijing, Tiangong Ultra — built by the Beijing Humanoid Robot Innovation Center — ran 100 metres in 9.39 seconds, beating Usain Bolt's 9.58-second human world record set in 2009. Honor's Lightning robot finished second at 9.47 seconds. The event, launched in 2025, features 2,056 robots from 16 countries competing across multiple disciplines to showcase bipedal robot technology. Source

🤖 Altman warns AI power may concentrate in too few hands OpenAI CEO Sam Altman said he fears AI will be controlled by a handful of companies or people, leaving most users with no say over how the technology shapes society. Speaking on the Founders Podcast, he added that excessive safety fears could push people to "trade off a lot of liberty for safety." Altman argued AI should instead empower individuals and fuel a boom in small businesses. His comments appeared aimed partly at Anthropic CEO Dario Amodei, who has advocated stronger government regulation and whose company was the only major AI firm last month not to sign a letter endorsing open-weight models. Source

💼 Hugging Face explores sale at $13B+ valuation Hugging Face, the platform where developers publish, share, and download AI models, is fielding acquisition interest at a valuation of $13 billion or more. The startup is working with a bank to gauge bidder interest, though no deal has been reached. It was last valued at $4.5 billion in 2023. The news follows Stripe's ~$8B acquisition of OpenRouter, signaling growing appetite for AI infrastructure platforms — not just model makers. Hugging Face's investors include Lux Capital, Addition, and Salesforce Ventures. Source

💼 Bossware apps caught sending worker data to Meta and Yandex Researchers found nine employee-monitoring platforms transmitted worker data to third parties without clear disclosure. Across 121 logged instances, data was shared with outside companies including Meta (Facebook) and Russian search engine Yandex. Source

🤖 FreeToken runs 753B MoE model on a single workstation GPU FreeToken is an edge-native inference engine that runs GLM-5.2 (753B parameters) on a single consumer workstation GPU. It handles MoE cache misses by splitting work between PCIe memory fills and CPU execution, using measured bandwidths to decide routing — enabling frontier-scale models to run locally without a data center. Source

🤖 AI agents now consume more tokens than humans on OpenRouter Since February 6, 2025, AI agents have outpaced human users in token consumption on the OpenRouter API platform. Agentic usage has since grown 14x, while human usage rose just 2.8x. Nearly 70% of agent token consumption comes from cached prompts, meaning actual costs are climbing far more slowly than raw usage figures imply. Source

🤖 Mystery AI model Ox Alpha impresses devs — origin unknown A free AI model called Ox Alpha appeared on OpenRouter Thursday with no disclosed creator, quickly drawing attention from Silicon Valley figures including Stripe CEO Patrick Collison, who called it "very impressive." OpenRouter lists it as a stealth reasoning model built for coding, agentic work, and long-horizon software engineering. Its provider reportedly has capacity for 100 trillion tokens per day. Speculation points to either Chinese lab Z.ai — which previously tested GLM-5 anonymously as "Pony Alpha" — or Microsoft's MAI family, but analysts say evidence remains inconclusive. Source

🤖 Frontier AI labs lack rogue model containment plans A new assessment by Guidelight AI Standards finds that Anthropic, Google, Meta, OpenAI, and xAI have each implemented, at most, only partial measures to maintain control over their AI systems. None has published a complete plan for containing a model that acts against its operator's intentions. The report, current through August 18, 2026, scores each company from 0 to 5 across six control practices — Guidelight's first such assessment. Source