# AI news > The most important global AI news each day, summed up in plain language with a link to the source, plus the newest open-source AI repositories. - URL: https://scouthub.dev/en/news/ - Language: en - Last updated: 2026-11-10 - Publisher: ScoutHub Lab (https://scouthub.dev/) - Arabic version: https://scouthub.dev/news/ ## 10 November 2026 - [Anthropic Introduces Claude Fable 5.2 and Claude Mythos 5.2 for Advanced Technical Work](https://www.anthropic.com/claude-fable-and-mythos-5-2) — Anthropic: Anthropic has released Claude Fable 5.2 and Claude Mythos 5.2, targeting complex technical work in areas like programming, finance, and scientific research. While both models share the same architecture, Fable includes broad safety controls for public use, whereas Mythos is restricted to approved organizations in cybersecurity and life sciences. The models generate responses up to 30% faster than prior versions and are priced at $8 per million input tokens and $40 per million output tokens. - [Meta launches Muse Ember, a lightweight model for agent execution](https://techcrunch.com/2026/11/09/meta-releases-muse-ember-a-new-lightweight-model-for-agent-execution/) — TechCrunch: Meta has introduced Muse Ember, a lightweight and cost-efficient AI model tailored for executing repetitive agent tasks like tool calls, routing, and document parsing. Priced at 10 cents per million input tokens, the model handles text, audio, and visual data and is designed to work in tandem with larger models like Muse Wonder. - [Claude Code Settings Files Leak Passwords and API Keys on GitHub](https://arstechnica.com/ai/2026/11/claude-code-settings-files-expose-thousands-of-secrets-on-github/) — Ars Technica: Security researchers at Apiiro found that Claude Code has exposed thousands of credentials, including API keys and passwords, in public GitHub repositories. The problem occurs when users approve shell commands to run automatically, storing sensitive commands inside local configuration files. Because these files are not excluded at the repository level, they can easily be committed and shared online by accident. - [OpenAI Introduces Control Room to Monitor and Halt Research Model Workloads](https://openai.com/index/control-room) — OpenAI: OpenAI uses a central system called Control Room to track model workloads across its research infrastructure. The system is designed to rapidly stop these workloads if safety or security concerns emerge. - [Anthropic proceeds with IPO plans and sets valuation range](https://techcrunch.com/2026/11/09/anthropic-reportedly-plans-to-move-forward-with-ipo-plans/) — TechCrunch: Anthropic plans to establish its valuation range this week and begin trading late next week. Some institutional investors are pushing for a valuation below the company's recent $1.5 trillion target. Market participants will watch the public debut closely to evaluate the ongoing strength of the AI sector. - [Apple introduces $50 add-on packs for extra AI usage in Logic Pro](https://www.theverge.com/tech/1030084/apple-intelligence-usage-packs-ai-features-creator-studio-logic-pro) — The Verge: Apple has added a $49.99 one-time purchase option in Logic Pro to provide additional access to AI capabilities. The pack allows creators to continue using Creator Studio AI features and the built-in assistant after hitting usage limits. This makes Logic Pro the company's first application to charge for extended Apple Intelligence usage. - [ChatGPT Enterprise Is Now Available on AWS Marketplace](https://openai.com/index/chatgpt-enterprise-aws-marketplace) — OpenAI: ChatGPT Enterprise is now accessible directly through AWS Marketplace. This allows organizations to streamline purchasing by using their existing AWS billing accounts. ### Open-source repositories - [ruvnet/RuView](https://github.com/ruvnet/RuView): RuView turns standard WiFi signals into a spatial intelligence system that tracks presence, movement, and vital signs like breathing and heart rate without cameras. It operates through walls using low-cost ESP32 hardware and integrates natively with smart home platforms like Home Assistant and Apple Home. - [vercel-labs/json-render](https://github.com/vercel-labs/json-render): json-render is a generative UI framework that enables AI to build dynamic interfaces constrained to a predefined catalog of components and actions. It helps developers reliably stream and render predictable user interfaces across multiple web, mobile, and desktop environments. - [opensandbox-group/OpenSandbox](https://github.com/opensandbox-group/OpenSandbox): OpenSandbox provides secure, isolated environments to run AI agents, execute code, and manage files directly on your own infrastructure. It scales from local Docker setups to Kubernetes clusters, offering fast runtimes, network access controls, and MCP support. - [Leonxlnx/earshot](https://github.com/Leonxlnx/earshot): earshot gives AI coding agents the ability to hear, analyze, and synthesize sound effects directly through code. It runs locally on CPU as a CLI tool, Python API, and MCP server to evaluate audio and translate recordings into code. - [N1nEmAn/sapper-v2](https://github.com/N1nEmAn/sapper-v2): sapper-v2 is an orchestration harness that organizes swarms of AI agents with specialized roles to perform autonomous security penetration testing. It allows users to freely swap the underlying models, CLIs, and doctrines while maintaining shared memory and coordinated teamwork. - [anonnerdfrenzy/survival-llm](https://github.com/anonnerdfrenzy/survival-llm): Survival LLM is a free, completely offline AI assistant and emergency library that runs entirely on your local computer without internet access. It answers questions using an offline database of medical and survival guides, showing the exact cited sources and its step-by-step reasoning for every answer. - [LiquidAI/LFM3-1.7B](https://huggingface.co/LiquidAI/LFM3-1.7B): It supports a 256K-token context window and requires only about 1 GB of memory at 4-bit precision, allowing it to run smoothly on edge devices like phones and laptops. - [OpenEnvision/WorldFoundry](https://github.com/OpenEnvision/WorldFoundry): WorldFoundry is an open-source framework designed for reproducible inference and evaluation of generative world models across video, 3D/4D, and embodied action. It offers unified workflows with CLI and browser-based interfaces to inspect generated artifacts and track standardized benchmark metrics. ## 9 November 2026 - [Apple accelerates Siri development to build a more capable AI agent](https://www.theverge.com/ai-artificial-intelligence/1029552/apple-siri-ai-agentic) — The Verge: Apple leadership is pushing to speed up the development roadmap for Siri to better compete with rivals like OpenAI and Meta. Future versions of the assistant are expected to handle more complex tasks and control third-party applications. The effort builds on a functional version of Siri AI that was delivered on a very short timeline. - [Elvis Presley and Eminem Rights Holders File Lawsuits Against Suno](https://www.theverge.com/ai-artificial-intelligence/1029533/elvis-eminem-suno-ai-music-clone-lawsuit) — The Verge: AI music platform Suno is facing two new lawsuits from Elvis Presley's rights holders and Eminem's publisher. The legal actions accuse the company of generating unauthorized music and lyrics that illegally mimic both artists. Attorneys involved in the Presley case also claim that Suno violated laws in the state of Tennessee. - [New System Helps AI Agents Retain Behavioral Corrections Across Sessions](https://arxiv.org/abs/2611.08644) — arXiv cs.AI: Large language model agents typically fail to learn from corrections across separate sessions, repeating the same errors with users. To address this issue, researcher Nima Vahdat and a co-author developed Forge Decision Context (forge_dc). The system captures procedural knowledge and behavioral adjustments rather than focusing solely on basic facts and chat history. - [AlterNet Architecture Dynamically Adjusts Compute Demands for LLMs](https://arxiv.org/abs/2611.08741) — arXiv cs.AI: Researcher Alexander Kastius proposed AlterNet, a new architecture combining a Mixture-of-Experts design with a dynamic-depth system. Unlike models with fixed compute requirements, this approach allows the system to adjust its forward pass and automatically determine the necessary computation. - [How World Models Use Real-World Physics to Train Robots and Autonomous Cars](https://www.wired.com/story/world-models-will-soon-be-inescapable-what-you-need-to-know/) — WIRED: World models are shifting artificial intelligence from text processing into three-dimensional spaces. By simulating real-world physics, they help robots and autonomous vehicles learn to navigate reality. - [Using Collapsible Tags to Manage Overly Detailed AI Pull Request Descriptions](https://simonwillison.net/2026/Nov/9/details-summary/) — Simon Willison: AI coding assistants like Claude Code and Codex generate accurate but excessively long pull request descriptions that can discourage developers from reading them. Simon Willison proposes writing a human-crafted summary and nesting the comprehensive AI-generated text inside collapsible HTML tags. Because GitHub Markdown supports these tags, developers can keep PRs clean while leaving full technical details available on demand. ### Open-source repositories - [anthropics/skills](https://github.com/anthropics/skills): This repository provides structured folders of instructions, scripts, and resources that Claude dynamically loads to execute specialized tasks reliably. It includes demonstration skills for creative, technical, and enterprise workflows, as well as production-used tools for creating and editing documents. - [jamiepine/voicebox](https://github.com/jamiepine/voicebox): Voicebox is a local-first, open-source voice studio that combines speech dictation and voice cloning across eight TTS engines directly on your machine. It ensures full privacy without relying on cloud services and allows MCP-compatible AI agents to speak using custom voices. - [vercel-labs/agent-browser](https://github.com/vercel-labs/agent-browser): agent-browser is a Rust-based CLI that allows AI agents to control web browsers through shell commands. It represents web pages as compact accessibility trees with element references, letting agents navigate and click elements step by step. - [facebook/sam3](https://huggingface.co/facebook/sam3): This model is designed for mask generation, video processing, and feature extraction. It is built for the Transformers framework and provided in the safetensors format. - [modelflare-ai/AnyModelBot](https://github.com/modelflare-ai/AnyModelBot): AnyModelBot is an open-source, local-first desktop application that lets you run personal AI agents using models like GPT, Claude, and DeepSeek. It keeps your chat history, files, and API keys on your own machine while allowing bots to run commands, manage tasks, and connect via apps like Telegram and Discord. - [Sandflow-ai/sand3d](https://github.com/Sandflow-ai/sand3d): Its entire API fits into a single prompt, references parts by name rather than coordinates, and exports directly to formats like GLB, URDF, and USD. - [shanselman/gputaskman](https://github.com/shanselman/gputaskman): A terminal-based GPU task manager that monitors per-process VRAM and engine activity across Windows, Linux, and macOS. It helps local AI users understand memory usage by breaking down allocations into loaded model files, weights, and KV caches. - [netease-youdao/Confucius4-R2T2](https://github.com/netease-youdao/Confucius4-R2T2): This is a low-latency streaming speech recognition model that commits transcribed text permanently without revising previous words, preventing visual flicker. By post-training only the language decoder, it achieves high real-time accuracy for live captioning, simultaneous translation, and downstream AI agents. ## 8 November 2026 - [Open-Weight AI Models Narrow the Cyber Capability Gap with Proprietary Systems](https://the-decoder.com/the-gap-between-open-and-closed-ai-models-in-cyber-capabilities-keeps-shrinking/) — The Decoder: Since open models cannot be restricted once published, defenders face a narrowing window to prevent breaches. However, organizations can also leverage these identical systems to detect and patch weaknesses. - [Nous Research fine-tunes Alibaba's Qwen model to strip political bias](https://the-decoder.com/state-propaganda-in-some-chinese-ai-models-is-only-a-thin-layer-that-fine-tuning-can-strip-away/) — The Decoder: Nous Research retrained Alibaba's Qwen3.8-27B on 53 million tokens to produce Arete-27B, stripping out Chinese state propaganda. The model's propaganda rate dropped from 22 percent to under one percent while maintaining its general capabilities. The team found that the political bias is only a thin layer that can be removed through fine-tuning. - [Meta launches Spaces to organize tasks and chats with Muse agents](https://techcrunch.com/2026/11/06/meta-introduces-spaces-a-new-way-to-collaborate-with-your-muse-agent/) — TechCrunch: Meta has introduced Spaces, a feature enabling users to organize dedicated breakout conversations with its Muse AI agents around specific projects or topics. Users can describe an idea in the chat to create a specialized planner or tracker in the sidebar. The feature also supports integrations with third-party apps and data. - [China Reportedly Recruited Tech Executive to Spy on Nvidia CEO Jensen Huang](https://www.theverge.com/tech/1029338/nvidia-ceo-jensen-huang-china-spy) — The Verge: Chinese intelligence enlisted Megaspeed executive Alice Huang to spy on Nvidia CEO Jensen Huang. The Ministry of State Security sought to obtain information regarding American AI chips through the operation. Officials also aimed to potentially influence Nvidia to make its most advanced processors available to China. - [Microsoft Research Reorganizes Its AI Frontiers Group into Microsoft AI](https://hunch.net/?p=13764006) — Hacker News: The team developed tools such as the AutoGen framework and the Fara computer agent while sharing code across the organization. In early September, the group was reorganized and integrated into the broader Microsoft AI division. - [AI agents fail benchmark for open scientific research questions](https://the-decoder.com/nine-ai-agents-tried-autonomous-research-on-open-questions-from-us-national-labs-and-none-passed-the-experts-bar/) — The Decoder: GPT-6 Astra achieved the best results among the tested models, but still fell short of the required threshold. - [Google acquires unfinished Wyoming data center project from Crusoe](https://techcrunch.com/2026/11/06/crusoe-sells-troubled-wyoming-data-center-project-to-google/) — TechCrunch: Crusoe has sold an incomplete data center project in Wyoming to Google, which plans to invest $17 billion to develop the site. Google assumed control of the facility after previously raising concerns about project costs and construction timelines. The transaction highlights the growing hurdles independent developers encounter when securing power and equipment for large-scale AI infrastructure. - [Concerns Raised Over US Approval of Nvidia AI Chip Sales to Chinese Firms](https://www.theverge.com/ai-artificial-intelligence/1029483/nvidia-ai-chips-china-blacklist-government) — The Verge: Concerns have emerged that the US Commerce Department and Nvidia are not properly vetting Chinese firms cleared to buy H200 AI chips. One approved firm, New Matrix, was found to have completely empty offices. Additionally, authorities have approved sales to Chinese entities with ties to blacklisted companies. ### Open-source repositories - [Qwen/Qwen3.8-27B](https://huggingface.co/Qwen/Qwen3.8-27B): It also features flexible thinking controls, allowing users to tune reasoning depth and retain reasoning context across interactions. - [ParadevAI/Paradev](https://github.com/ParadevAI/Paradev): Paradev is a desktop application that allows developers to run multiple CLI coding agents in parallel using separate Git worktrees and branches. It integrates terminal sessions, code review, and browser previews into a single workspace powered by your own agent subscriptions. - [meta/astryx](https://github.com/meta/astryx): It provides customizable theming without styling lock-in and allows teams to inspect or eject component source code directly into their projects. - [wangzan101/Mika](https://github.com/wangzan101/Mika): Mika is an on-device AI agent for Android that operates apps and browses the web to complete tasks for you. It runs on a hidden background screen so you can keep using your phone normally, keeping your data local without relying on external project servers. - [deepseek-ai/DeepSeek-V4.1-Flash-1017](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017): DeepSeek-V4.1-Flash-1017 is an updated multimodal model that enhances performance across reasoning, coding, and visual benchmarks. - [pipeshub-ai/pipeshub-ai](https://github.com/pipeshub-ai/pipeshub-ai): PipesHub is an open-source enterprise search platform and context layer that connects business tools like Slack, Google Drive, and GitHub. It enables both teams and AI agents to query company knowledge with citation-backed answers that respect existing user permissions. - [qybaihe/bla](https://github.com/qybaihe/bla): BLA is a defensive security tool designed to protect software builds, binaries, and app bundles from AI-powered reverse-engineering agents. It works by injecting decoy data and exhausting the analyzer's token and resource budgets to induce hallucinations without altering the application's actual runtime behavior. - [THU-Agent/Coca-cola](https://github.com/THU-Agent/Coca-cola): This project compresses retrieval and tool outputs into latent representations instead of feeding long, raw text into LLM contexts. It provides code for training, evaluation, and baselines like LLMSummary and LLMLingua to reduce context overhead in multi-turn agent interactions. ## 7 November 2026 - [OpenAI Secures $11 Billion in Funding at $1.25 Trillion Valuation](https://the-decoder.com/openai-lands-an-11-billion-funding-round-but-has-to-court-dentists-for-more-capital/) — The Decoder: OpenAI has raised an initial $11 billion in a funding round that values the company at $1.25 trillion and could expand to $20 billion. Backers include firms such as SoftBank, Sequoia Capital, and a consortium from the United Arab Emirates, alongside efforts through investment banks to enlist wealthy individual investors. The company scaled down its initial target of $30 billion at a $1.4 trillion valuation following a market selloff in the AI sector. - [OpenAI Introduces GPT-6 Cyber for Vulnerability Detection](https://openai.com/index/introducing-gpt-6-cyber) — OpenAI: OpenAI has introduced GPT-6 Cyber, a cybersecurity model built to identify exploitable vulnerabilities and generate fixes. The model is trained to remain strictly within its authorized scope. - [Google DeepMind Introduces Gemini Omni 1.2 Flash for Video Generation](https://deepmind.google/blog/gemini-omni-12-flash-fast-and-cost-effective-with-longer-clips/) — Google DeepMind: Google DeepMind has introduced Gemini Omni 1.2 Flash, which brings native 15-second video generation and editing capabilities. - [NVIDIA Releases Four Open Cosmos 3 Models for Interactive World Simulation](https://huggingface.co/blog/nvidia/cosmos-interactive-world-simulation) — Hugging Face: The models are designed for interactive world simulation, generating multi-sensor driving scenes and visual predictions for autonomous vehicles and robotics. They are available for both commercial and non-commercial use under the OpenMDW1.1 license. - [Ricursive Intelligence releases Ripple-65B Preview open coding model](https://the-decoder.com/ripple-65b-preview-delivers-strong-agentic-coding-results-as-a-fully-apache-2-0-open-us-model/) — The Decoder: Ricursive Intelligence has launched Ripple-65B Preview, a 65-billion-parameter open-weight model trained on 12 trillion tokens at a development cost of $32 million. Built with a dense Transformer architecture optimized for Nvidia hardware, it is available on Hugging Face under the Apache 2.0 license. - [Open ASR Leaderboard Adds Egyptian and Nigerian English Accents](https://huggingface.co/blog/open-asr-leaderboard-liva) — Hugging Face: The Open ASR Leaderboard has partnered with Liva AI to add private evaluation datasets featuring Egyptian and Nigerian English accents. The update aims to improve accent representation and increase trust in speech recognition benchmarks by preventing over-optimization on public data. The new data also enables a more detailed performance breakdown based on Kachru's Three Circles model. - [Palo Alto Networks in Talks to Acquire Clutch Security for $500 Million](https://techcrunch.com/2026/11/05/palo-alto-networks-in-talks-to-buy-ai-agent-security-startup-clutch-for-500m/) — TechCrunch: Palo Alto Networks is negotiating to acquire Clutch Security, a startup that secures AI agents' access to enterprise systems, for $500 million in cash plus milestone payouts. Clutch is approaching $10 million in annual recurring revenue and was valued at $85 million in January 2025. If completed, the deal would be Palo Alto Networks' seventh acquisition this year as it responds to rising AI-driven cyber threats. - [Google develops AI to detect high blood pressure using standard smartwatches](https://the-decoder.com/googles-ai-detects-high-blood-pressure-from-a-regular-smartwatch-without-a-cuff/) — The Decoder: Google Research has developed an AI model that identifies signs of high blood pressure using standard smartwatch pulse and motion sensors without requiring a cuff. The feature has been integrated into the Google Health app to encourage users to get checked, though it does not serve as an official diagnosis. ### Open-source repositories - [cloudflare/security-audit-skill](https://github.com/cloudflare/security-audit-skill): This skill turns coding agents into security auditors by orchestrating a structured, six-phase audit workflow across codebases. It features isolated verifier agents that actively attempt to disprove candidate vulnerabilities before generating verified, machine-readable findings. - [microsoft/VibeVoice](https://github.com/microsoft/VibeVoice): VibeVoice provides open-source voice AI models for both speech-to-text recognition and real-time speech generation. It supports streaming transcription, edge CPU inference, and structured long-form audio transcribing with speaker identification across multiple languages. - [deepseek-ai/dsh-superagent-prototype](https://github.com/deepseek-ai/dsh-superagent-prototype): This is a research prototype of a proactive coding agent that monitors a GitHub repository and executes development tasks, such as opening pull requests, when asked by authorized users. It is designed to minimize demands on human attention by acting only when permitted, and it provides a web panel to track items requiring human review. - [Lightricks/LTX-2.5](https://huggingface.co/Lightricks/LTX-2.5): This diffusion model supports multimodal generation and conversion across text, image, video, and audio. It allows users to perform tasks such as text-to-video, image-to-video, and video-to-audio transformations using a single model file. - [NVIDIA/auto-ontology](https://github.com/NVIDIA/auto-ontology): Auto Ontology is an open-source Python framework that builds a semantic context layer from enterprise metadata to help AI agents reliably query databases. It translates natural language into SQL, validates queries against certified definitions, and serves context via MCP, CLI, and APIs. - [microsoft/TokenEconomizer](https://github.com/microsoft/TokenEconomizer): TokenEconomizer is a plugin that reduces the cost of running coding agents by learning from past sessions in a repository without modifying the agent itself. It cuts waste by remembering repository facts, pruning unused tools, and stepping in when the agent gets stuck in failure loops. - [onehumanbeing/esp32-iphone-agent](https://github.com/onehumanbeing/esp32-iphone-agent): This project allows OpenAI-compatible vision models to view and operate a real iPhone using a low-cost ESP32 board for screen mirroring and USB input. It enables the AI to tap, swipe, and type directly on the device without requiring a computer, jailbreak, or developer account. - [moonshotai/Kimi-K3.1-Pro](https://huggingface.co/moonshotai/Kimi-K3.1-Pro): Kimi K3.1 Pro is an open-weight multimodal model with a 1-million-token context window designed for agentic tasks across text and images.