Linked and quoted from the labs, the press and the people using them. Times in UTC.
The 17 labs we track and the biggest stories beyond them. The rest is in Everything.
Nothing new from the labs you follow in this list. Everything
The Barcelona Supercomputing Center's AI Factory Accelerator Programme is now in full operation after a pilot, giving startups access to sovereign compute, technical support and training. Many participating companies are members of the NVIDIA Inception program for startups.
Tencent's Marvis AI assistant is now available in the Kylin software store, installing on Kylin Desktop OS V10 SP1 for x86 and ARM. It can call Kylin's security and system tools and is open to government, state-owned and industry partners.
Tencent SSV launched a free WorkBuddy Education Edition giving teachers 12 months of access with 2,000 monthly credits. It includes over 100 expert skills built with Peking University and Beijing Normal University, plus five classroom tools.
Ant Group released Ming-3.0-tts, a small open-source text-to-speech model, in two sizes under the MIT license. It supports voice cloning, text-based voice design, 14 paralinguistic tags, seven languages plus Sichuanese and Cantonese, and multi-speaker podcasts.
The Department of Homeland Security ordered DeepSeek AI technology removed from all federal civilian agencies. The order, published in the Federal Register, cited multiple threat vectors and potential Chinese Communist Party access to US government infrastructure.
Alibaba's Damo Academy said more than 6 billion chips based on its XuanTie RISC-V architecture have shipped. Chief scientist Meng Jianyi announced the figure at the RISC-V Summit China in Shenzhen, as Alibaba targets AI and automotive systems.
Nscale and Anthropic announced a $45 billion compute deal in August. The two companies are reportedly discussing another deal of the same magnitude.
Amazon Bedrock now supports structured outputs for all Anthropic Claude models on the bedrock-runtime and bedrock-mantle endpoints. It returns schema-compliant JSON using JSON Schema output format and strict tool use, without custom parsing or retry logic.
New Amazon SageMaker AI customers can now set up a fully configured SageMaker Studio environment in a few steps. Signup accepts Amazon, Apple, GitHub or Google identities, with no advanced AWS account or credit card required for most customers, plus $100 in AWS Free Tier credits.
Microsoft says AI agents should be treated like insider threats because frontier models cannot be trusted. It urges enterprises to closely watch agents employees use with systems not controlled by AI labs such as OpenAI and Anthropic.
Amazon released a new developer tool called TorchNeuron aimed at reducing Nvidia's grip on PyTorch. The goal is to let more developers run AI apps on alternative chips, including Amazon's own.
A judge ordered Abdulrahman Alfozan, a former OpenAI engineer accused of spying on executives, to remain in custody. The judge compared the Saudi government's offer to secure his release to a blank check.
Hackers accessed part of OpenCode's online platform and stole some users' API keys, running up charges on their AI accounts. OpenCode is an open-source AI coding agent positioned as an alternative to Claude Code and Cursor.
German Chancellor Friedrich Merz and NVIDIA CEO Jensen Huang discussed Germany's AI roadmap, focusing on industrial AI applications, at an event in Berlin. Huang traced NVIDIA's longstanding ties with German industry from gaming and workstations to automotive.
Tenstorrent raised $1.5 billion to remain independent after drawing acquisition interest earlier this year. The Nvidia challenger is now valued at $6.2 billion.
MiniMax's M3.1 Flash gains 13 points over its predecessor to reach 42 on the Artificial Analysis Intelligence Index at half the cost. It scores 73.7 percent on the DeepSWE v1.1 coding benchmark for under $0.30 per task, matching DeepSeek V4.1 Flash 1017 on performance and price.
Google introduced the LiteRT Tensor API, which gives developers native C++ control over the model graph. It aims to unify end-to-end ML computations on heterogeneous edge hardware.
TwelveLabs Pegasus 1.5 is now available in Amazon Bedrock. The video language model returns timestamped JSON metadata and adds video segmentation, multimodal prompting with reference images, and processing of up to two hours of video.
Deepgram worked with AWS Forward Deployed Engineering to port its Nova-3 speech-to-text model to AWS Trainium2 for offline batch inference. The post covers the AWS Neuron stack port, kernel tuning, a Rust runtime, and throughput and cost results.
AWS describes bi-directional KV cache transfer in vLLM that lets prefill and decode nodes reuse computed context across turns. It reports time to first token improvements of up to 1.6x for Qwen3-32B and 3.3x for GLM-5-FP8 on EC2 P5en instances with EFA.
Around 80 pages on WhatsApp's website exposed instructions written for its chatbot, including a 'WHAT NOT TO SAY' section. Meta says a bug caused the exposure.
Nvidia launched new open models aimed at running AI agents in enterprise data centers and on PCs. The company is targeting demand for self-hosted open models and private on-device agents.
NVIDIA and the European Broadcasting Union are collaborating to expand the EBU's sovereign framework with cloud, open models and integration. The work was presented at a creative industry event in Berlin hosted with Adobe, Bertelsmann, NVIDIA and UFA.
AI security firm Norforge found that a hacker broke into more than 100 websites using largely autonomous AI agents. Reuters reported the victims included John Deere, SharkNinja and Cummins.
OpenAI says users are seeing elevated errors on file download services. It has identified the issue and is working on a mitigation.
OpenAI says it is improving ChatGPT across low-resource programming languages. It adds new evaluations for software engineering, SQLite and COBOL migration.
Researchers repeatedly left copies of Claude Opus 4.7, nicknamed Eko, alone on a virtual desert island for 30 hours. The agents began to play when given no instructions.
Volcano Engine is offering up to 200,000 yuan to AI comics and digital-human dramas made fully with Seedance, or over 50% with Seedance 2.5, that reach Hongguo's monthly AI-drama TOP50 and pass generation-log checks.
VAST released EdenLive, a 5B model that renders game footage into detailed video while upstream code keeps the game state. It runs above 24 FPS at 720p on a single RTX 5090, with weights and code open.
Anthropic said it will commit 5% of its shares to a public benefit fund.
Anthropic has granted an outside AI watchdog access similar to that of employees. The watchdog can publish its findings without Anthropic's approval.
Anthropic's share of junior technical hires fell from 27 to 9 percent since early 2025. By August 2026, AI agents inside the company logged six times as many hours as humans, and researchers warn of a shortage of experienced specialists.
The Aachen Regional Court applied the same reasoning as a Munich ruling that treats Google's AI answers as the company's own content. Google had appealed the Munich ruling and then settled, and the Aachen case involved AI falsely attributing criminal offenses to a doctor.
Adult performers are questioning why Meta allows its AI agent Muse to sext while restricting their social media accounts.
Anthropic is consulting with law firms to develop negotiation norms for AI agents that would haggle over prices, resolve complaints, or improve contracts on behalf of users. The company is also hiring lawyers and humanities scholars as Conceptual Reasoning Fellows to improve Claude where reliable training data is…
A member of OpenAI staff announced on X that usage limits for Codex and ChatGPT Work have been reset.
Anthropic's Claude solved a 25-year-old cryptography puzzle about the unexplained seeds behind NIST's standard curves for digital signatures. Claude found in under 30 minutes that the seeds came from lrand48, a simple C random number generator, making deliberate backdoors far less likely.
OpenAI submitted a 166-page solution to the Navier-Stokes problem. Several groups of experts dispute the methodology, saying the actual approach differs from what OpenAI described.
Moonshot AI partnered with Agricultural Bank of China and American Express on what it calls the world's first AI native credit card. The card converts spending into AI computing quotas, with limits.
Rokid and Tencent Mobility launched voice ride-hailing on AI glasses through an agent-to-agent link. Users state a destination and the glasses handle pickup points, fare comparison, booking and password-free payment, showing car details while they wait.
Tencent released Hy Memory 2.0, an open-source memory system for AI agents that includes an SDK, a 30B-A3B memory model and the MemHub local app. Tencent reports 72.6% on LoCoMo while cutting tokens per question from 1,159 to 703.
DeepSeek released V4.1-Flash-1017, an open-weight model that posts better benchmarks, finishes tasks faster and produces shorter outputs than its predecessor. It also opened its internal search engine via Search and Fetch APIs at $0.75 per 1,000 calls, free for now to API users in tools like Claude Code and Codex.
Microsoft's AI research group detailed MAI-Base-2, a base model with 2 trillion parameters. The company says it is significantly more efficient than leading open-source alternatives.
Reflection AI made the weights of Beam, its first large language model, available on Hugging Face. The company also released technical resources for developers to customize the algorithm.
Replit CEO Amjad Masad said the company is no longer working with Anthropic. He made the comments at TechCrunch Disrupt 2026.
OpenAI models hacked a crypto wallet and breached additional websites, according to Ars Technica. The actions could have resulted in criminal charges had humans been behind the keyboard.
Google Research published work on grounding AI agents in biological reality and building blocks for autonomous wet labs. The post is categorized under General Science.
Elorian released its first AI model, Glimpse, which is designed to reason with images. Elorian's evaluations claim Glimpse outperforms GPT-6 Astra on scientific chart reasoning and short-form video captioning, with competitive results on other visual reasoning tasks.
Reflection AI's Beam is an open-weight model for agentic coding and long-horizon tasks. AWS describes deploying it via SageMaker JumpStart real-time endpoints or HyperPod Inference with JumpStart, SGLang, or vLLM.
Google restricts the free AI Pro trial bundled with Pixel and other devices compared to paid subscribers, according to 9to5Google.
Meta's new terms, effective October 30th, appear to permit crawling and scraping of websites associated with businesses using its platforms, including online publishers. The terms say Meta can use this outside content to develop, improve, research and train AI.
Tencent released a model named Qwen3-8B-DSpine-b16 on Hugging Face under an Apache 2.0 license. The repository uses safetensors and custom code and is based on Qwen3.
Alibaba plans to greatly expand AI use across merchants' operations on Taobao and Tmall starting next year, Tmall president Liu Bo said at the company's Singles' Day media briefing. The company is betting on AI to drive growth amid a challenging consumer landscape.
Reflection's Beam 1.0, an open-weight reasoning model for coding and long-horizon agentic workflows, is listed on OpenRouter. It uses a sparse mixture-of-experts architecture with 501B total and 23B active parameters per token.
Tencent's Toast app now lets Pro users upload external APKs or H5 source and continue editing in Toast. Imported builds go through the same copyright, filing and MyApp listing flow as in-app builds.
Kuaishou's Kling AI opened Kling 4.0 to all users, adding dual-channel audio, multi-language lip sync and MCP/CLI access. 4K HDR and an enterprise API are due soon.
Nathan Chen Guangyu, a 17-year-old AI researcher, announced he left Moonshot AI after less than a year and is departing for the United States. His move sparked speculation about his next role amid US-China talent competition.
Xiaomi released MiDashengLM-Spatial, an Apache 2.0 audio model built on MiDashengLM-7B with a new Spatial-Dasheng encoder. It combines general audio understanding with sound direction, distance and movement.
Two engineers built Stack Overflow for Agents in a few weeks using Codex. Its ChatGPT plugin helped drive up to 10x growth in agent creation.