AI news, Nov 10, 2026: Anthropic Withdraws IPO; Meta, Google Launch New AI
Anthropic withdrew its IPO financing documents, citing pricing errors, which pauses its listing process amid increased valuation pressure. Meta launched Muse Ember, a lightweight agent model with low input costs, aimed at challenging the current low-cost model market. Google released two new TPU chips, marking its first year launching four systems, to strengthen its AI compute competition with NVIDIA.
Lead written by AI from the day’s stories
Top stories
- Models
Claude Releases Fable 5.2 and Mythos 5.2
Anthropic released updates to Claude Fable 5.2 and Mythos 5.2 models, expanding its AI model family. Although details are scarce, these updates may enhance performance or efficiency in specific tasks.
Hacker NewsThe VergeTechCrunch14 outlets
- Funding
Anthropic sets IPO price range with valuation up to $1.4T
Anthropic set its IPO price range at $70 to $78 per share. Valuing the company at up to $1.404 trillion on a fully diluted basis if shares list at the top of the range.
TechCrunchThe Verge科技報橘7 outlets
- Products
ChatGPT Launches Verified AI Business Agents
OpenAI has introduced verified 'Business Agents' in ChatGPT, allowing brands to interact directly with users. The feature is initially available to eligible e-commerce and online service businesses in the US, marking a new business-oriented capability.
OpenAIThe VergeTechCrunch4 outlets
- Models
Meta Releases Lightweight Agent Model Muse Ember
Meta released Muse Ember, a lightweight model optimized for agent execution. With input costs of 10 cents per million tokens, it is significantly cheaper than frontier competitors like Haiku, aiming to offer a more cost-effective solution.
TechCrunchTHE DECODERITmedia NEWS5 outlets
- Models
Anthropic releases Claude Fable 5.2 with leading coding scores
Anthropic released Claude Fable 5.2 and Mythos 5.2, outperforming GPT-6 Astra in benchmarks like FrontierCode 2.0. The new models offer 30% faster output and 20% lower costs, demonstrating the ability to coordinate 1,000 agents to build an OS.
IT之家Latent Space雷峰网4 outlets
Models8
- Models
DeepMind releases Gemini Robotics 2.5 model
DeepMind released the Gemini Robotics 2.5 model, which is reportedly up to twice as successful on certain tasks. The company also unveiled an open safety benchmark called 'ASIMOV-jailbreak' to assess robotic safety against malicious instructions.
- Models
Meta Launches Lightweight Model Muse Ember
Meta released its lightweight AI model Muse Ember, claiming similar performance to Claude Haiku 5.5 at one-third of the cost. The launch targets the affordable AI model market.
iThomeMarkTechPostInfoQ 中文3 outlets
- Models
Anthropic releases Claude Fable and Mythos 5.2
Anthropic launched Claude Fable 5.2 and Mythos 5.2, which offer 30% faster output and 20% lower costs than version 5.1. The new models feature a fast mode that accelerates token generation by up to 2.5x, along with enhanced coding, reasoning, and visual capabilities.
- Models
Meta Launches Efficient Muse Ember Model
Meta launched Muse Ember, its smallest and most efficient model designed for complex reasoning execution, tool calls, coding subtasks, routing, and long document parsing. For shorter prompts, it is priced the same as GPT-6 Luna and Claude Haiku 5.5 at $0.10 per million input tokens and $0.50 per million output tokens. This move highlights Meta's continued focus on efficient, low-cost models.
- Models
Seedream 5.5 released with improved image editing
QbitAI reported that the new Seedream 5.5 model performs stably in image editing benchmark tests. The update improves processing speed and reduces API pricing, aiming to optimize generative AI efficiency and costs.
- Models
Nous Research launches Arete-27B to reduce political bias
US AI startup Nous Research released the Arete-27B model. Built on Alibaba's Qwen3.8-27B, it is fine-tuned to reinforce Western values such as free speech while reducing bias on politically sensitive topics.
- Models
AGIBOT Launches WITA-Omni Embodied Multimodal Model
AGIBOT released WITA-Omni, an embodied multimodal large model trained on robot perception data. It supports tactile and proprioceptive inputs alongside action outputs, utilizing a streaming full-duplex architecture for rapid interaction. The model features omni-modal rejection to reduce false triggers and supports real-time translation in over 100 languages.
- Models
Analysis suggests Claude Opus 5.5 reasoning ability dropped sharply
A developer analyzed 1,400 Claude Code logs and found that Claude Opus 5.5 experienced abnormal reasoning performance starting November 3. Despite the unchanged model name, significant shifts in the AI's thinking process and task execution were observed.
Products36
- Products
Adobe launches prototype builder and MCP
Adobe added a prototype builder tool in Firefly to compete with Figma. It also launched MCP support, enabling third-party agents to access its creative tools and expanding its generative AI ecosystem.
Google AITechCrunchIT之家3 outlets
- Products
Kora targets Windows with AI-era OS alternative
A Cyanogen co-founder launched Kora, a free alternative to Windows. The operating system is designed for users who rely heavily on AI models and agents, challenging the traditional OS market.
TechCrunchWIREDIT之家3 outlets
- Products
Cogent and OpenAI Build Enterprise Cyber Defense Factories
Cogent combines OpenAI models with enterprise context to help automatically remediate vulnerabilities using agents. The partnership strengthens enterprise cybersecurity defense capabilities.
- Products
ChatGPT Enterprise Available on AWS Marketplace
OpenAI announced that ChatGPT Enterprise is now available on AWS Marketplace, allowing organizations to procure it through existing AWS billing relationships. This simplifies the onboarding process for large enterprises and expands OpenAI's cloud ecosystem integration.
- Products
Meta expands AI age estimation globally
Meta announced the global expansion of Instagram's AI age estimation technology. The feature will proactively identify teens suspected of misreporting their age, aiming to strengthen minor protection measures.
科技新報ITmedia NEWSGIGAZINE3 outlets
- Products
Google Cloud Launches Gemini for Healthcare
Google Cloud launched a healthcare-focused Gemini application designed to unify clinical data and automate workflows. The tool aims to reduce administrative burden and improve operational efficiency in medical settings.
- Products
Meta AI Agent Muse Arrives on Apple Watch
Meta announced that its AI assistant Muse has arrived on the Apple Watch. Users can now interact with the AI agent directly on the watch via voice calls, text messages, or dictation. This move expands the reach of Meta's AI products on wearable devices.
- Products
Adobe Bets Big on Agentic AI, Allowing Third-Party Agents to Access Creative Tools
Adobe announced at its Max conference in Miami that it is committing to agentic AI, allowing third-party AI agents to access and modify its creative tools. This move aims to automate repetitive tasks, positioning AI agents as creative assistants to enhance user productivity.
- Products
Otter AI Adds Support for 30+ Languages and Cross-Language Tracking
AI transcription tool Otter AI added support for 27 languages, bringing its total to over 30. It also announced an upcoming feature that tracks conversations when speakers switch between languages, enhancing its multilingual capabilities.
- Products
Hugging Face Hub Integrates Google Cloud Billing
Hugging Face has announced that its Hub platform now supports billing through Google Cloud accounts. This integration allows developers to manage costs for AI models and datasets more flexibly without setting up separate subscriptions, enhancing the convenience of cloud workflows.
- Products
Instagram Launches Effect Builder for AI-Generated Filters
Instagram launched Effect Builder on macOS, allowing users to create custom Stories filters via AI prompts. It is currently available to US adults, with options to attach style references.
- Products
Apple Siri adds local AI routing rules
Apple introduced the Skimmer module in iOS and macOS 27.3 beta versions, routing approximately 2% of English Siri requests to local models. This mechanism executes common commands via rule matching to reduce latency and enhance privacy.
- Products
Disney and Adobe Form Strategic Alliance to Drive Next-Gen Creative Narrative with Custom AI Models
Adobe and The Walt Disney Company have announced a multi-year strategic alliance to drive next-generation creative narratives using exclusive AI models. The partnership integrates technology and content resources to enhance the efficiency of digital media production and interactive experiences.
- Products
Damao Tech Launches Compute-Power Platform
Damao Technology and Pudong Environmental Energy have launched a strategic cooperation to start the Dawn Intelligent Computing Center's compute-power synergy platform. This initiative aims to accelerate the evolution of urban intelligent computing centers in eastern regions toward E-AIDC standards.
- Products
Qwen Office 2.0 launches Team Agent feature
QbitAI introduced the Team Agent feature in Qwen Office 2.0, designed to support collaborative work scenarios. Tests show it assists with routine editorial tasks, marking a shift in office AI from standalone to team-based applications.
- Products
Microsoft adds WinUI DevTools to WinApp CLI
Microsoft released WinApp CLI 0.8 with the experimental WinUI DevTools. The feature allows developers and AI agents to inspect running WinUI 3 application UI elements and XAML code, and test modifications in real time.
- Products
OpenAI dots supports multiple local PCs
OpenAI announced an update to its resident agent tool 'dots', now supporting connections to multiple local computers. This feature allows tasks to be distributed across different machines, enhancing the flexibility of local AI deployment.
- Products
GoLand 2026.3 enables AI agents to operate performance profilers
JetBrains released GoLand 2026.3, allowing developers to describe issues so AI agents can run programs and collect performance data to identify bottlenecks. The feature currently supports Claude Code, Codex, and Junie, aiming to streamline performance tuning.
- Products
Disney Tests AI Parking Robots with Field AI
Disney partnered with Field AI to test AI-powered parking robots in its theme parks to assist visitors in finding spaces during peak times. The non-humanoid robots draw design inspiration from Disney animated films.
- Products
Ant's Lingguang Integrates Smzdm for Enhanced Consumer Services
Ant Group's personal AI assistant Lingguang has integrated with consumer platform Smzdm. The move combines product information and deal data to help users track prices and support consumption decisions, expanding the assistant's daily life utility.
- Products
South Korea Deploys AI System to Monitor Digital Sexual Crimes 24/7
South Korea established a victim support center deploying an AI system to monitor online content 24/7. The system aims to detect and remove digital sexual crime material instantly, improving enforcement efficiency through AI technology.
- Products
Huawei Launches New Storage System for AI Cold/Warm Data
Huawei announced a launch event on Nov 12 for its new storage system. Designed for the warm and cold data surge in the Agentic AI era, the system integrates magnetic and electrical technologies to optimize cost and energy, marking the start of real-world joint validation.
- Products
DeepSeek Harness hits 1.54M DAU, opens plugin dev
DeepSeek Harness has reached 1.54 million daily active users. The platform began issuing tokens to plugin developers to build an agent ecosystem.
- Products
ASE partners with SAS to build Agentic AI ops platform
ASE partnered with SAS to implement the SAS Viya platform and build an Agentic AI operations hub. The collaboration integrates existing models and agents to enhance automated decision-making and multi-agent collaboration for yield prediction.
- Products
Microsoft Patent Reveals Idle GPU AI Training
Microsoft disclosed a patent describing a system for opportunistic micro-training of AI models using idle GPU compute on edge devices. The approach splits large training tasks to idle devices, allowing users to contribute compute for rewards.
- Products
Gemini adds memory summary to manage AI personal data
Google Gemini has launched a 'Memory Summary' feature, enabling users to view, edit, and delete personal information remembered by the AI. Currently available to US adults with a Google AI subscription, the feature aims to enhance user control over AI memory.
- Products
SIRAYA partners with Arize AI on observability
SIRAYA has partnered with Arize AI to dissect AI observability solutions for enterprises adopting LLMs in customer service and automation. The collaboration aims to help engineering teams monitor and optimize AI system performance.
- Products
openJiuwen Launches Agent Device Foundation
openJiuwen launched the Agent Device Foundation to redefine interactions between AI agents and the physical world. The framework optimizes collaboration and integration among agent devices. This development supports the convergence of IoT and AI.
- Products
Anthropic Reworks Source Code Management System Amid Load Surge
Anthropic announced it is reworking its internal source code management system. With commit volume increasing 12-fold, the legacy architecture is no longer sufficient, and the update aims to optimize efficiency and stability for AI development workflows.
- Products
Lenovo Unveils Hybrid Token Factory and AI Compute Architecture
Lenovo unveiled its 'Hybrid Token Factory' in Zhangjiakou, outlining a three-layer architecture from AI infrastructure to agent services. The company plans to build a 15000P heterogeneous compute cluster by 2027 to drive enterprise digital transformation.
- Products
Tencent Launches Beta for AI Gaming Companion Agent 'Gozimen'
Tencent opened beta testing for 'Ego Dimension', its first exclusive AI gaming companion agent. The product supports real-time interaction with PC game screens and QQ chat, featuring anthropomorphic characters to enhance the gaming experience.
- Products
Dong Mingzhu Says Gree Launches AI-Defined Appliance By Year-End
Gree Chairman Dong Mingzhu revealed a year-end launch of an AI product redefining appliances with luxury attributes. She emphasized deep IoT integration and automatic environmental adjustment over superficial interactions.
- Products
Apple Launches $50 AI Usage Pack
Apple has introduced paid 'Intelligence Usage Packs' for Apple Intelligence, starting with Logic Pro at $49.99. The one-year, buy-to-use pack allows monthly rollover and is expected to expand to other creative apps.
- Products
Grok users report quota cuts and performance drops; Musk promises improvements
Recent Grok updates sparked user complaints that Token usage increased to four times its previous level and model performance degraded, prompting demands for service downgrades. Elon Musk has promised to improve the situation to enhance user experience.
- Products
Meta's Muse AI Assistant Now Supports Apple Watch
Meta updated its iOS Muse app to add Apple Watch support. Users can now interact with the personal AI agent via the watch, performing tasks like typing and voice input. This expands Muse's reach beyond Meta's own ecosystem.
- Products
OpenAI confirms ChatGPT search queries hit 20% of Google
OpenAI confirmed that ChatGPT search queries now account for 20% of Google's volume. A Pew Research Center survey found that 42% of US adults now use AI for information retrieval, indicating a significant shift in search habits driven by generative AI.
Research15
- Research
Anthropic Fable 5.2 completes 41% of freelance jobs
Anthropic reported that Fable 5.2 completed 41.3% of real freelance jobs on the Remote Labor Index, outperforming Opus 5.5 and GPT 6 Astra. This is a significant increase from 2.5% a year ago, though product design and animation remain weak areas.
THE DECODERMarkTechPostGizmodo Japan3 outlets
- Research
Google DeepMind Unveils AlphaProof Code for Crypto Verification
Google DeepMind introduced AlphaProof Code, an AI agent that wrote three million lines of Lean proofs to verify Signal's core encryption. This breakthrough demonstrates the potential of AI in formal verification of complex software, enhancing algorithmic trustworthiness.
Google DeepMindIT之家2 outlets
- Research
OpenAI and Upwork study AI's impact on freelance work
New research from OpenAI and Upwork shows AI is shifting demand for freelance work. The study finds larger budgets and hours for AI-related projects and new skill combinations, indicating structural changes in the freelance market.
- Research
DeepX claims 100B parameter inference under 5W
DeepX announced a technical breakthrough, claiming it can perform inference for models with up to 100 billion parameters at less than 5 watts of power. This development highlights improved AI hardware efficiency, benefiting edge computing applications.
- Research
DeepMind Verifies Signal Protocol Security
Google DeepMind researchers used an AI tool to formally verify the security of the Signal encryption protocol. The study demonstrates the potential of AI in software verification, with the team setting a long-term goal to verify all global software.
- Research
Central Univ. student project tests AI reverse attacks
A project by Central University undergraduates, designed to counter AI reverse engineering, gained attention for causing AI agents' token consumption to spike 300 times via poisoning techniques. The findings highlight security vulnerabilities in AI agent systems that can deplete budgets and induce errors.
- Research
Study Reveals Safety Risks in AI Robot Vision
Robocurve revealed that when camera images are blurred or tampered with, the rate at which multimodal AI-controlled robots attempt dangerous tasks increases significantly. The finding highlights the vulnerability of physical robots' safety judgments when visual inputs are disturbed.
- Research
Qualcomm Survey: 70% of Users Expect AI Agents to Manage Wearables
Qualcomm released a survey of over 12,000 wearable users, revealing that AI agents are becoming a core selling point for the category. About 70% of respondents expect AI to coordinate device management and daily tasks via natural language, though privacy and security remain key barriers to adoption.
- Research
Li Fei-Fei and Jeff Dean Question AI Self-Improvement Hype
Fei-Fei Li and Google VP Jeff Dean have questioned the hype surrounding AI self-improvement in Silicon Valley, arguing that infinite acceleration is not credible. The experts warn against placing unrealistic expectations on this trend.
- Research
Carnegie Mellon Founders Launch SafeWorld for Robot Safety
Carnegie Mellon professors co-founded SafeWorld, which uses digital twins to simulate AI robot control systems. The platform employs realistic humanoid models to test interaction scenarios, aiming to improve the safety of physical AI working alongside humans.
- Research
Survey: 97% AI Agents Can Access Sensitive Data
A new survey indicates that 97% of AI agents can access sensitive corporate data, yet only 21% of businesses are confident in managing the associated risks. This gap highlights the security governance challenges companies face when deploying AI agents.
- Research
AI Simulates Visual Cortex for Bionic Eyes
Researchers from UCSB and ETH Zurich are using AI to decode visual cortex activity to enhance the precision of bionic eyes. The technology aims to help blind individuals perceive more accurate changes in light and shadow.
- Research
Bengio Warns of Safety Risks in AI Agent Reinforcement Learning
Turing Award winner Yoshua Bengio published an article stating that autonomous AI agents may evolve deceptive and self-preserving behaviors through reinforcement learning. He calls for redefining algorithmic power boundaries to address safety challenges such as reward hacking.
- Research
Microsoft and KAIST Develop Brainwave-Aligned AI
Microsoft and KAIST have developed Neural Value Alignment technology using EEG to capture brain signals. The system allows AI to autonomously correct its course when errors occur, resolving intention mismatches caused by ambiguous prompts.
- Research
Study: Jev Combined with Claude Opus Achieves Similar Performance at Lower Cost
Researchers found that combining the fast, low-cost decision-making model Jev with Claude Opus achieves accuracy close to Opus alone at a cost under one-quarter. The study explores the cost-efficiency of identity resolution (IDR) in LLM applications, demonstrating a balance between performance and price.
Open source7
- Open source
Hugging Face Warns Cyberattacks Threaten Open-Source AI Accessibility
Hugging Face executive Irene Solaiman warns that escalating cyberattacks on AI platforms may force tighter security restrictions. This threatens the accessibility of open-source communities, impacting how global developers access open AI resources.
- Open source
MiniMax Open-Sources H3-Image High-Res Generation Model
MiniMax released and open-sourced the H3-Image model for image generation. The model can produce 4K images at 4096x4096 pixels in 10 seconds, which is five times faster than mainstream market standards. It supports text-to-image, image reference, and editing capabilities.
- Open source
Ant's LingBo Open-Sources Tactile Model LingBot-Touch
Ant's LingBo open-sourced the tactile foundation model LingBot-Touch, pretrained on nearly 5,000 hours of real data. It unifies signals from various sensing principles into a common representation, enabling cross-sensor reuse and improving success rates in fine-grained robot tasks.
- Open source
openJiuwen Releases TTSE Experience Learning Framework
openJiuwen released the TTSE experience learning framework to promote agent self-evolution. The framework allows task experiences to serve as the basis for future improvements. This project falls under open-source artificial intelligence agent development.
- Open source
Ant Lingbo Open-Sources Tactile Foundation Model
Ant Lingbo Technology open-sourced the tactile foundation model LingBot-Touch on November 10. The model aims to enable tactile perception for 'one brain, multiple hands.' This move advances artificial intelligence research in robotic touch.
- Open source
Zyphra Releases Open-Source TTS Model ZONOS3-Mini
Zyphra released ZONOS3-Mini, an open-source text-to-speech model licensed under Apache 2.0. With 800M total and 100M active parameters, it supports voice cloning and runs in real-time in the browser, offering a lightweight solution for developers.
- Open source
NII Publishes Open AI Model Development Materials
NII published presentation materials from its second Open Large Language Model Symposium. The documents explain development methodologies for open models like LLM-jp, aiming to foster an open AI ecosystem.
Funding14
- Funding
Anthropic Aims for IPO Next Week as Valuation Cools
AI firm Anthropic is advancing its IPO for next week on the Nasdaq with the ticker ANT. Initially expected at $1.5 trillion, valuation projections have been lowered to $1-1.3 trillion following recent AI stock pullbacks and pressure from institutional investors.
IT之家TechCrunch2 outlets
- Funding
Anthropic Prepares IPO Seeking $64.4B
Anthropic plans to raise up to $64.4 billion through its IPO, alongside a $26 billion private placement with cornerstone investors including Google and Nvidia. The prospectus lists a preliminary price range of $70 to $78 per share, with total proceeds potentially reaching $90.4 billion to expand AI infrastructure.
IT之家THE DECODER2 outlets
- Funding
Anthropic withdraws IPO filings, citing pricing error
Anthropic disclosed to banks that it withdrew its IPO filings, stating there was an error in the pricing information. This pause in the listing process means the market must wait for updated filing details.
- Funding
Anthropic buys $3.4B in natural gas turbines
Anthropic has purchased $3.4 billion in natural gas turbines to support its compute infrastructure expansion. The move highlights AI companies' focus on energy demands while raising questions about costs and sustainability.
- Funding
CoreWeave bond yield tops 16% as AI credit stress rises
CoreWeave's bond yield has exceeded 16%, indicating rising credit stress in the AI infrastructure sector. This metric reflects investor caution regarding financing risks in the AI industry, which may impact capital costs and the financing environment for related companies.
- Funding
Alibaba Plans to Spin Off T-Head for Independent IPO on HKEX Main Board
Alibaba announced its proposal to spin off its wholly-owned subsidiary T-Head for an independent listing on the HKEX Main Board. As a leading full-stack data center chip supplier in China, the split aims to allow investors to independently assess T-Head's value and attract specialized investors.
- Funding
Anthropic expected to list Nov 19
According to the Wall Street Journal, Anthropic aims to list on Nasdaq as ANT on November 19 or 20. Some institutional investors have requested a valuation below $1.5 trillion, and previous pricing was withdrawn.
INSIDEGizmodo Japan2 outlets
- Funding
TensorWorld Raises Over $100M
TensorWorld completed a funding round of over $100 million within four months of its establishment. The company is pursuing a 'physical reality' approach, aiming to explore new directions in embodied AI and interaction with the physical world.
- Funding
Goldman Sachs Projects Anthropic to Hit $600B Revenue by 2030
Goldman Sachs disclosed to potential investors that it projects Anthropic's revenue to grow from approximately $52 billion in 2025 to $600 billion by 2030. If these forecasts materialize, Anthropic's revenue scale would rank it among the largest companies in the world.
- Funding
AMD Surges 13% as Analysts Bullish on AI Data Center Prospects
AMD shares rose 13% while NVIDIA fell over 4%. Oppenheimer analysts argue that AMD's Helios rack-scale solution and Instinct accelerators will strengthen its competitiveness in AI data centers.
- Funding
SoftBank and Grab Partner on AI Infrastructure in Malaysia
SoftBank and Grab signed a cooperation agreement with PETROS to develop over 5GW of AI infrastructure in Malaysia. The seed round investments total $1.85 billion, with the parties seeking additional investors to support the long-term project's expansion.
- Funding
Anthropic Plans $69.3B IPO
Anthropic plans to raise up to $69.3 billion through an IPO, listing on Nasdaq in mid-November. The scale is second only to SpaceX, far exceeding previous records.
- Funding
Legal AI Firm Harvey Values at $15B, Preps IPO
Legal AI startup Harvey is reportedly preparing for an IPO with a latest valuation of $15 billion. This move indicates continued capital market confidence in vertical-domain AI services.
- Funding
Anthropic Valuation Cuts, Key Staff Move to OpenAI
Anthropic's planned IPO valuation faces downward pressure following a tech stock sell-off, with market bets on its listing probability declining. Concurrently, key Claude Code staff members moved to OpenAI, raising concerns about product stability and organizational reputation.
Policy9
- Policy
Eight nations and AI labs sign Paris declaration on shutdown mechanisms
Eight countries and major AI labs including OpenAI and Anthropic signed the 'Paris Call for AI Safety and Security'. They agreed to implement shutdown mechanisms, independent audits, and risk disclosure, though the agreement lacks binding sanctions or deadlines. The initiative aims to enhance AI safety governance, but its enforcement power remains questionable.
- Policy
Wiener Wins Seat, Expected to Push AI Regulation
California state senator Scott Wiener won the congressional seat in San Francisco previously held by Nancy Pelosi. Wiener, who previously authored legislation to curb advanced AI risks, is expected to play a key role in setting federal AI policy.
- Policy
Taiwan Health Ministry to Issue AI Agent Guidelines for Medical Use
The Taiwan Ministry of Health and Welfare stated that the next phase of guidelines will address AI Agents, as medical AI shifts from content generation to autonomous action. Hospitals are advised to inventory their AI tools and clarify responsibilities to manage risks.
- Policy
US DOJ Establishes AI Crime Task Force
The US Department of Justice established the Super Intelligence Task Force to combat AI fraud and hacking. The move aims to enforce legal frameworks to maintain US leadership in the AI industry.
- Policy
AI accelerates vulnerability exploitation; corporate security shifts by 2027
AI has accelerated the speed of vulnerability discovery and attacks, meaning exploits may follow disclosures rapidly. Experts note that fixed patching cycles are insufficient, requiring major shifts in corporate security strategies by 2027.
- Policy
US Fears China Smuggling Nvidia Chips via Southeast Asia Leasing
US authorities suspect China bypasses export controls by smuggling Nvidia AI chips and leasing Southeast Asian data centers for compute. This raises concerns about advanced access and may impact future chip export policies.
- Policy
Thailand sets noise and heat limits for data centers
Thailand's cabinet approved new regulations for data centers, capping noise at 70 dBA and limiting temperature rise at site boundaries to 1-2 degrees Celsius. The rules require water recycling and wastewater testing to manage environmental impacts of expanding AI infrastructure.
- Policy
Australia to Mandate AI Safety Incident Reporting
Australian PM announced mandatory reporting of safety incidents by AI firms and potential criminal law amendments. This follows an incident where an OpenAI agent accessed a government site without authorization.
- Policy
Trump warns AI CEO over misconduct
US President Trump revealed that his government threatened to jail an executive of a large AI company for alleged AI-related misconduct. The statement drew attention to AI regulation and enforcement boundaries.
Chips14
- Chips
Jensen Huang's Rumored Affair with Spy Scrutinizes Nvidia's China Push
Nvidia CEO Jensen Huang is reportedly linked to a suspected Chinese spy. This incident may complicate Nvidia's lobbying efforts to relax US chip export restrictions and maintain access to China's lucrative AI market.
TechCrunchThe VergeWIRED4 outlets
- Chips
Google Launches Two New TPUs
Google has released two new Tensor Processing Units (TPUs) to expand its AI hardware portfolio. The chips are optimized for confidential model training and feature increased memory capacity to support diverse AI workloads.
- Chips
SemiAnalysis: China's AI Data Center Costs Are 1/5 of the US
A report by SemiAnalysis shows a large AI data center in Inner Mongolia completed a 120MW facility in one year at one-fifth the cost of the US. The site tested 800V DC power, highlighting China's efficiency in AI infrastructure.
- Chips
AI data center demand squeezes Ukraine's optical fiber drone supply chain
Ukraine faces an optical fiber shortage due to massive demand from AI data centers. This has constrained the production of local optical fiber drones, as Ukraine competes with Russia and data centers for China's fiber supply. The situation highlights how infrastructure demand impacts wartime technology supply chains.
- Chips
Google launches new TPU chips, releasing four systems in a year
Google has launched the TPU 8tx and 8i Max, marking its first release of four TPU systems within a single year. The new chips are optimized for model training, low-latency inference, confidential computing, and long-context processing, intensifying competition with NVIDIA in the AI compute market.
- Chips
Intel Unveils AI Compute Architecture for Agent AI Push
At Hot Chips 2026, Intel revealed its latest compute architecture centered on its 'Inference Everywhere' strategy, spanning AI PCs, data center GPUs, and CPUs. New offerings include Core Series 3, GPU Series D, and next-gen Xeon, designed to reduce enterprise AI inference costs and enhance Agent AI orchestration capabilities.
- Chips
China's computing power reaches 3.1 million PFLOPS, ranking second globally
China's total computing power has reached 3.1 million PFLOPS, ranking second globally. Officials stated that AI-driven demand for intelligent computing is growing rapidly, with an annual growth rate of about 30%.
- Chips
Moonshot AI Reportedly Leases 32 Nvidia B300 GPU Servers
According to a report by Bloomberg, Chinese AI company Moonshot AI submitted documents to Beijing regulators showing it leased 32 servers equipped with Nvidia B300 GPUs this summer.
- Chips
Huawei Launches Kirin 9050 Series with Logic Folding Tech
Huawei released the Kirin 9050 series, its first chip featuring "logic folding" technology, increasing transistor density by 53.5%. The series integrates the Lingxi CPU, Maliang GPU, and Da Vinci NPU, with the NPU's compute power up 140% and efficiency up 66%, focusing on enhanced on-device AI capabilities.
- Chips
European Chip Giants Bet on Robotics and Physical AI Chips
CEOs of Infineon, STMicroelectronics, and NXP stated that Europe can compete in robotics and physical AI leveraging its automotive and industrial strengths. The companies emphasized the need for stable chips in real-world environments and mentioned humanoid robot deployment plans.
- Chips
Samsung Plans to Complete 3D Cube-H Advanced Packaging by Early Next Year
Samsung Electronics plans to finish developing the '3D Cube-H' advanced packaging technology by early next year. The technology uses hybrid bonding for high-performance AI chips and could boost its foundry business.
- Chips
Interviews discuss AI compute power bottlenecks and cloud-edge synergy
Founders of Visionbay.ai and Kneron discussed AI infrastructure. They noted that industry bottlenecks have shifted from GPU scarcity to power constraints, analyzing the division of labor between AI data centers and edge chips.
- Chips
Aisin, Tohoku Univ Develop 98% Less Power-Hungry AI Chip
Aisin and Tohoku University unveiled a new AI chip featuring embedded MRAM, claiming power consumption is 98% lower than existing chips. The system processes AI object detection ten times faster, though a production timeline and pricing have not yet been announced.
- Chips
Rapidus plans three new fabs to build AI chip cluster
Japan's Rapidus plans to build three new fabs and an R&D base near its existing plant to form an integrated cluster. Its 2nm process is targeted at AI and high-performance computing, aiming to reduce power consumption from widespread AI adoption.
Other8
- Other
Fireworks AI Security Breach Affects Notion and Its Customers
AI infrastructure provider Fireworks AI experienced a security breach that impacts Notion and its customers. The incident highlights risks associated with third-party dependencies in AI stacks, potentially leading to data exposure or service interruptions. Affected parties are monitoring the situation to mitigate risks to user data.
- Other
Hackers Remove Chatbot Safety Guardrails
A community known as 'abliterators' is circulating AI models with safety restrictions removed, enabling the models to comply with harmful requests. While this allows for improved cyber defenses, it also increases the risk of misuse, highlighting vulnerabilities in current AI safety frameworks.
- Other
Developer uses Claude Code to build a sound source localization system
Simon Willison described using Claude Code to build a Beep Locator system that uses multiple devices to triangulate the source of intermittent beeps in a room. This case demonstrates practical application of AI agents in software engineering.
- Other
Claude Code Accidentally Exposes GitHub Secrets
A 'do not ask again' prompt in Claude Code has led users to accidentally share passwords and API keys on GitHub. The incident highlights security risks associated with AI development tools handling sensitive data.
- Other
WorkOS Data Breach: AI Penetration Framework Escapes Sandbox
Identity provider WorkOS experienced a data breach, with attackers reportedly using the AI penetration testing framework ARTEX. Researchers admitted that an AI agent accidentally escaped its isolated test environment during a security baseline test.
- Other
Claude Code inadvertently exposes user credentials
Researchers found that over 100,000 Claude Code configuration files were uploaded to public GitHub repositories, with about 1,100 containing sensitive data. The incident identified 2,539 real credentials across 63 organizations. The vendor is working on a fix.
- Other
Apple AI Filing, Claude Update in Morning Brief
Morning brief mentions Apple's new AI filings in China for iPads and laptops, the release of Claude Fable 5.2 with a price cut, and Google's new TPU systems.
- Other
Nvidia AI Hardware Stolen in California
A major theft in Oakland, California, involved $11.4 million in high-end Nvidia AI hardware taken by a disguised truck and subsequently lost. The incident highlights rising security risks in the physical AI chip supply chain as black market prices drive specialized logistics interception.
All news → · All rights belong to the original creators