中文 ↗RSS ↗

THE NEWSROOM / AI 新闻

AI News · Page 4

Source-linked briefings on models, agents, products and practical AI.

YYSI publication dates appear on the left; original announcement dates appear under each brief.

Search and categories filter this page. Use the page links below for earlier news.

REPORTED

Product updates OpenAI

OpenAI Launches Verified 'Business Agents' in ChatGPT Beta

According to The Verge, OpenAI is beta-testing verified 'Business Agents' in ChatGPT that represent brands and can answer questions like comparison shopping, The Verge reports. Eligible US e-commerce and online service businesses can create agents via a Business Hub, which also offers verification and insights into ChatGPT mentions. Chats with these agents are labeled and kept separate from the main conversation. OpenAI says this is an early step toward deeper business participation.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates Anthropic

Anthropic Launches Claude Meetings, Whiteboards and Sheets for Workplace Collaboration

According to TechCrunch, Anthropic announced Claude Meetings, which joins video calls as a visible participant to draft plans, map timelines and pull data from connected apps in real time, plus Claude Whiteboards and Claude Sheets for collaborative visual workspaces and spreadsheets. All three launch in beta on paid plans, with Meetings limited to Team and Enterprise on Google Meet. Claude Data is now generally available.

AI summary · Media reportOriginal source ↗Read brief →
SOURCE

Product updates NVIDIA

NVIDIA Donates OpenShell Agent Security Runtime to CNCF, Unveils DSX Kubernetes Tools

At KubeCon + CloudNativeCon, NVIDIA contributed its OpenShell agent security and governance runtime to the CNCF Sandbox and backed Google's Agent Substrate for scaling agent workloads on Kubernetes. It also announced DSX AI Factory Platform additions: NVIDIA Kubernetes Engine and Kubernetes Recipes in public preview, plus open-source NVIDIA Snapshot, which restores inference replicas in seconds without model reload.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Models & research Google

Google and Artificial Analysis launch Multimodal Knowledge Agent benchmark

Google DeepMind and Artificial Analysis released MKA, a 200-task benchmark testing how agents reason over visually complex files across 10 verticals and 83 occupations. Tasks take experts three to eight hours and are graded by an autorater that agreed with human experts 96% of the time. Google says Gemini 4 Carbon ranked second in quality at about 6x lower cost, and is available via the Gemini API.

AI summary · Official sourceOriginal source ↗Read brief →
SOURCE

Models & research Google

Google releases Gemini 4 Carbon for developers with frontier cybersecurity and long-horizon agentic capabilities

Google announced that developers can now build with Gemini 4 Carbon, which it describes as its strongest model, with frontier performance in cybersecurity and long-horizon capabilities. Google says Carbon is built for sustained agentic workflows spanning software engineering, full-stack web development and scientific research. It is available through the paid API tier in Google AI Studio and Android Studio, and Google AI Ultra subscribers can try it in AI Studio and the Antigravity platform.

AI summary · Official sourceOriginal source ↗Read brief →
REPORTED

Product updates Meta

Meta Brings Muse AI Agent to Apple Watch and Wear OS

According to TechCrunch, Meta announced that its Muse AI agent is now available on Apple Watch and Android's Wear OS, letting users interact with the assistant via voice calls, text messages, or dictation on their smartwatch. The expansion follows Muse's September App Store debut and a dedicated iPad app, as Meta pushes the consumer assistant deeper into its device ecosystem.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Models & research MiniMax

MiniMax Open-Sources H3-Image Model for 4K Image Generation

According to IT Home, MiniMax released H3-Image, an open-weight image generation model derived from H3, supporting text-to-image, image reference and image editing. According to the company, it can produce a 4096x4096 (about 16MP) image within 10 seconds, claims up to 16 reference images and stronger instruction following than other open models. Weights are on Hugging Face, with an API and a first-week discount on 16MP requests.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Models & research Google

DeepMind unveils Gemini Robotics 2.5 and ASIMOV-Jailbreak safety benchmark

According to Ars Technica, Google DeepMind revealed Gemini Robotics 2.5, a vision-language-action model it says boosts task success 30-100% and speed 20-30% over the prior version in lab tests. Gemini-powered Apptronik Apollo 2 robots kitted automotive parts in European Mercedes-Benz factories with a 96% success rate. DeepMind also released ASIMOV-Jailbreak, an open-source benchmark of 1,500 prompts designed to test whether robotic AI refuses unsafe instructions.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates AI industry

Adobe Adds AI Agents and Local MCP to Creative Cloud Apps

According to TechCrunch, At its MAX event in Miami, Adobe announced AI features across Photoshop, Premiere, Illustrator, InDesign and Firefly, including a proactive always-on agent and a local Model Context Protocol server exposing Creative Cloud capabilities to chatbots such as ChatGPT, Claude and Gemini. Firefly gains a private-beta prototyping tool called Builder, an AI video assembly tool and on-brand graph creation. Adobe said it is still deciding which MCP endpoints to expose.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates AI industry

Spotify Expands AI DJ With New Voices, Personalities and a Low-Chatter Mode

According to TechCrunch, Spotify is upgrading its AI DJ, used by over 94 million Premium listeners, with selectable voices and personalities in English and Spanish, plus an option to cut commentary between songs. The new voices begin rolling out November 10 to Premium users in the more than 95 markets where DJ is supported. Spotify says the low-chatter setting suits background listening like parties.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Models & research Google

Google DeepMind uses AI agents to formally verify Signal encryption protocol

According to IT Home, Google DeepMind researchers used autonomous AI agents to formally verify the Signal end-to-end encryption protocol, producing a mathematical proof of its security properties, IT之家 reports. The method also flagged known vulnerabilities in older Signal implementations. The analysis took about four weeks and required substantial compute; Google told Wired it is unclear whether the approach scales broadly. The work verifies the protocol, not the full Signal app.

AI summary · Media reportOriginal source ↗Read brief →
SOURCE

Product updates Google

Google Cloud launches Gemini for healthcare in preview with HCA, Highmark and Northwell

Google Cloud announced Gemini for healthcare, a solution that connects electronic health records, billing platforms, clinical guidelines and Medicare datasets to automate multi-step clinical and administrative workflows such as records synthesis and prior authorization review. The company says it operates within defined guardrails with human oversight. It is initially available in preview with launch partners HCA Healthcare, Highmark Health and Northwell Health.

AI summary · Official sourceOriginal source ↗Read brief →
REPORTED

Models & research AI industry

Agibot Releases WITA-Omni Embodied Omni-Modal Interaction Model

According to IT Home, Agibot announced WITA-Omni, which it calls the industry's first embodied omni-modal end-to-end interaction model. The vendor says it adds tactile and proprioceptive inputs plus robot action and facial-expression outputs, runs on a streaming full-duplex architecture with roughly 400-600 ms response latency, supports wake-word-free interaction and streaming interpretation across many languages, and can proactively initiate exchanges. These are company claims.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates AI industry

SoftBank and Grab Sign Pact With PETROS to Develop Malaysian AI Infrastructure

According to IT Home, SoftBank and Southeast Asian ride-hailing platform Grab signed a cooperation agreement with Sarawak oil company PETROS to develop AI infrastructure in Sarawak and Peninsular Malaysia, IT之家 reported. The parties plan seed investments of $1 billion and $750 million for the two projects, with PETROS contributing $100 million, and will seek other investors and hyperscale data-center customers for a multi-year project with a design scale exceeding 5GW.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Practical guides AI industry

Thailand Approves Environmental Rules for Data Centers Amid AI Investment Boom

According to IT Home, Thailand's cabinet approved new environmental rules for data centers, Bloomberg reported via IT之家. Uninterruptible power systems must keep noise near 70 dBA, site-boundary temperature rise is capped at 1-2°C, and cooling water must be recycled with wastewater tested. Industrial-estate facilities face extra requirements. Officials said projects failing safety and environmental standards will not proceed, as AI-related projects make up about 70% of over $50 billion in investment applications this year.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates AI industry

Caltech-DARPA Robot Autonomously Assembles Antenna Truss in Orbit

According to IT Home, A Caltech-built robot funded by DARPA's NOM4D program assembled a third truss section of a 1.4-meter reflector antenna in orbit, which researchers call the first autonomous structural build in space. Ground teams still handled planning and commands through brief daily communication windows. DARPA says the demo is a step toward building structures kilometers in size.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Models & research AI industry

MIT-Led PLOW Team Wins Up to $53M ARPA-H Award to Build Generative AI for Safer Herbicides

According to MIT News, MIT will lead an up-to-$53 million ARPA-H STREAM award for PLOW, a project developing generative AI models that predict chemical compounds or siRNAs targeting the same plant enzymes as glyphosate while optimizing for low resistance risk, low synthesis cost and low human and ecological harm. Partners include the Broad Institute, UNC, Tennessee, Illinois and Robigo; candidates will undergo high-throughput screening.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Models & research AI industry

Ant Group's Lingbo Open-Sources LingBot-Touch Tactile Foundation Model for Robots

According to IT Home, Ant Group's Lingbo unit open-sourced LingBot-Touch, a tactile foundation model pretrained on nearly 5,000 hours of real tactile data. It unifies magnetic, vision-based and piezoresistive sensor signals into one representation reusable across hands, sensors and action models. In 26 real-robot tests, adding it raised average task success by roughly 14 to 26 percentage points across four action models.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates AI industry

Boston Dynamics Tests Wheeled Atlas Humanoid as Research Platform

According to IT Home, Boston Dynamics released video of its Atlas humanoid fitted with standard electric-scooter wheels via custom adapters on its ankle actuators, with no major hardware rebuild and only a simulated foot-to-wheel swap in software. The wheeled Atlas can jump and spin and sprint laterally, letting engineers stress actuators during acceleration, turning and stopping. Boston Dynamics says it is a research platform, not a commercial product.

AI summary · Media reportOriginal source ↗Read brief →
REPORTED

Product updates AI industry

Tencent Opens Beta for 'Eciyuan' AI Game Companion Agent

According to IT Home, Tencent has opened a beta test for "Eciyuan," described as its first dedicated AI game-companion agent. According to IT之家, the PC-focused product offers real-time voice chat, can view the player's game screen and answer questions about it, and supports companion chat through PC and mobile QQ. It features four characters covering titles such as Valorant, League of Legends and Counter-Strike; desktop-pet and voice-change features are listed as coming soon.

AI summary · Media reportOriginal source ↗Read brief →