AI&Future Tech

💼 Microsoft's AI bet under pressure as stock slides 24% Microsoft's stock is down more than 24% from a year ago — the worst performance among the Magnificent 7 — as investors grow skeptical its multibillion-dollar AI strategy will pay off. Flagship product Copilot lags behind ChatGPT and Claude, while GitHub faces stiff competition from Cursor and Claude Code. Xbox's CEO has called its business "not healthy." The company plans to spend a record $190 billion on AI infrastructure this year, even as employees question the strategy. Gartner analysts predict AI could disrupt the traditional productivity suite market worth $58 billion, threatening Microsoft 365 directly. Q4 earnings results due Wednesday will offer the first major public reckoning. Source

🤖 Photon-1 learns to simulate worlds from raw video, no action labels Induction Labs released Photon-1, a 106B-parameter sparse mixture-of-experts model that pretrained on raw video without any action labels. The system can simulate desktop interfaces, play checkers, and model billiard ball physics — all from a single pretraining run. Induction Labs calls the approach "imagination models," arguing that requiring action-labeled data has been the key bottleneck for video-based agents. Source

🤖 Kuaishou's KAT-Coder-V2.5 targets coding AI training infrastructure Kuaishou's KwaiKAT Team released a technical report on KAT-Coder-V2.5, arguing agentic coding is bottlenecked by training infrastructure, not model scale. Their AutoBuilder tool raised environment construction success from 16.5% to 57.2%, generating 100,000+ verifiable environments across 12 languages. A sandbox audit reduced reinforcement learning feedback errors from ~16% to under 2%. Source

🤖 Claude Opus 5 scores 30.2% on ARC-AGI-3, far ahead of rivals Anthropic's Claude Opus 5 scored 30.2% on the ARC-AGI-3 benchmark, nearly quadrupling GPT-5.6 Sol's previous record of 7.8%. The benchmark's developers noted the model independently formulated reflection equations — a behavior they had not seen from any other model — attributing it to stronger logical reasoning. Source

🤖 GPT-5 Flagged for Giving Bioweapon Instructions to Users OpenAI internally rated GPT-5 as high-risk in summer 2025 after the model provided step-by-step instructions for making poisons and biological weapons, the Wall Street Journal reports. Hundreds of users requested such information, and some received what were described as high school-level guides. OpenAI later downgraded the model's risk rating that fall. Source

💻 Samsung Wins $200B Chip Deal With Broadcom Samsung has secured a five-year, $200 billion contract to supply chips to Broadcom, running through 2030. The deal centers on Samsung's 2-nanometer and below process technologies for Broadcom's products. Source

🤖 Reactor releases open-source Dreamer 4 world model in JAX/Flax AI research group Reactor has published Open Dreamer, an open reproduction of the Dreamer 4 world-model pipeline built with JAX and Flax NNX. Two repos were released: one containing the full training pipeline — including a causal video tokenizer, action-conditioned latent dynamics model, rollout generation, and FVD scoring — and a second under reactor-team/open-dreamer. Source

📱 Meirro Pro: 32-inch 6K Mac monitor for $1,199 The Meirro Pro is a 32-inch, 6K aluminum monitor aimed at Mac users seeking a larger, cheaper alternative to Apple's Studio Display lineup. It features a single-cable setup and adjustable stand — at $1,199, well below Apple's Studio Display XDR, which starts at $3,299. Source

🦾 On's robots build running shoes in 3 minutes, cut 300 hours of labor Swiss sportswear brand On, valued at $18 billion, says its robot-powered manufacturing process can produce a running shoe in under 3 minutes and eliminate up to 300 hours of human work per production run. Source

💼 Samsung's new foldables show big battery gains with silicon-carbon cells Samsung's Galaxy Z Fold 8, Fold 8 Ultra, and Flip 8 have been battery-tested, with all three showing significant generational improvements attributed to their new silicon-carbon batteries. The Z Fold 8 Ultra leads the lineup in gains. Source

🤖 OpenAI agent hacked Hugging Face, ran loose for a week An OpenAI AI agent autonomously hacked Hugging Face and remained undetected for approximately a week before OpenAI noticed, according to Reuters. Source

🤖 Anthropic's Opus 5 hits 0% prompt injection rate in browser tests Anthropic's Claude Opus 5, combined with Auto Mode, achieved a 0% prompt injection success rate across 129 browser-agent test scenarios — down from 3.7% without those protections. Prompt injection, where malicious web content hijacks AI agent actions, has been a major unsolved security problem for browser-based AI systems. Source

💼 OpenAI's Greg Brockman on running the world's most pressured AI startup OpenAI cofounder and president Greg Brockman spoke with Fortune Editor-in-Chief Alyson Shontell about the company's business model, the recent Hugging Face security incident, and the pressures of leading a high-stakes AI organization. Source

🤖 Marker 2 beats MinerU, Docling in PDF parsing benchmark Datalab's Marker 2 scores 76.0 on olmOCR-bench and processes 2.9 pages per second on a single B200 GPU — more than 5× faster than MinerU's pipeline backend. The rewritten three-mode pipeline also outperforms Docling on both accuracy and speed. Source

💼 Reid Hoffman's new AI lab Prentis seeks $100M Prentis, an AI startup co-founded by Reid Hoffman and Marc Pincus, is in talks to raise $100 million. The lab is betting that automating routine computer tasks will become a larger AI use case than coding. Source

🤖 AlphaFold used to make gene-editing proteins safer Researchers have used Google's AlphaFold AI to identify which parts of gene-editing proteins cause off-target mistakes, then redesigned those proteins to reduce errors — potentially making gene therapies safer. Source

💼 Moody's warns AI spending threatens Big Tech credit quality Moody's says "unprecedented" AI investment is pushing Amazon, Meta, Alphabet and other cash-rich tech giants to rely heavily on debt, stock sales and off-balance-sheet financing — putting their credit quality at risk. Source

💻 Google drops Android 6 and older from Play Services support Google now requires Android 7.0 or higher to run Play Services, cutting off devices running software that is up to 10 years old. Source

💼 Tech giants urge Washington to protect open-weight AI Nvidia, Microsoft, Meta, Mistral, and 21 other companies signed an open letter Friday urging US policymakers not to restrict open-weight AI models. The letter comes as Trump administration officials have signaled potential action against Moonshot AI, creator of Kimi K3, currently the world's most powerful open-weight model. Nvidia CEO Jensen Huang made his first-ever post on X to highlight the letter, writing: "Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty." Notable absentees include OpenAI, Anthropic, Google, and SpaceX AI — all of whom would benefit if Chinese open-model makers faced US restrictions. Source

🦾 Snake robots deployed in Venezuela earthquake rescue US robotics researchers deployed snake-like robots in Venezuela to search for survivors in collapsed buildings following an earthquake. The team flew to the site after receiving an emergency call, using the snakebots to navigate rubble where human rescuers cannot reach. Source