AI news — 18 November 2026

Every morning we collect the most important AI news from dozens of sources and sum it up in a few lines, with the newest open-source projects beside it. Each card links to the original source.

RSS · Today's newsRSS · New repositories

News 7 stories

OpenAI
Models & releasesCovered by 3 outlets

OpenAI Introduces GPT-6.2 Astra and Sol

OpenAI has introduced GPT-6.2 Astra and Sol, offering improved professional outputs, coding, and computer use capabilities. The new models also provide higher efficiency at a lower cost across ChatGPT Work, Codex, and the API.

Original

Introducing GPT-6.2 Astra and Sol

Introducing GPT‑6.2 Astra and Sol, with better professional outputs, coding, computer use, and efficiency at lower cost across ChatGPT Work, Codex, and the API.

Read at the source Also covered by: Simon Willison، Hacker News
Google AI
Tools & productsCovered by 3 outlets

Google makes Tasks in Gemini free in six countries

Google has made Tasks in Gemini available free of charge across six countries. The tool includes deeper integration with Google Workspace to complete actions on behalf of users, alongside new capabilities debuting in the U.S.

Original

Put Gemini to work for you

Tasks in Gemini is now free of charge in six countries, with new features starting in the U.S. and deeper Google Workspace integration to get things done on your behalf.

Read at the source Also covered by: TechCrunch، The Verge
Ars Technica
Tools & products

Microsoft integrates Office apps directly inside Copilot

Microsoft announced that Word, Excel, and PowerPoint are now generally available inside Copilot, turning the AI assistant into the primary workspace. Users can create and edit documents, spreadsheets, and presentations directly within Copilot using context from emails, chats, and meetings. The company also introduced Copilot Autopilot, an autonomous agent designed to handle broader ongoing responsibilities across productivity tools.

Original

Copilot is "the new OS for work," and now Office runs inside of it

Word, Excel, and PowerPoint now live inside Copilot; always-on agents are next.

TechCrunch
Policy & safety

FBI warns that hackers are using open-source AI tool Artex to breach networks quickly

The FBI warned that cybercriminals are utilizing Artex, an open-source AI-powered penetration testing tool, to autonomously compromise networks in under three hours. Attackers connect the tool to large language models to rapidly exploit security flaws, having already targeted customers of US firm ConnectWise and financial institutions in South Korea.

Original

FBI says Artex AI open source hacking tool used to breach networks in ‘under three hours’

The FBI says that the open source hacking tool was used in cyberattacks targeting a U.S. IT giant and the South Korean financial industry.

Ars Technica
Policy & safety

Google adds separate toggle to stop Gemini from training on user data

Google is updating Gemini settings to separate conversation history controls from AI model training. Users can now retain their past chats while opting out of letting the company use their prompts and attachments for training. The opt-out choice applies only to future interactions, leaving previously shared data in the training set.

Original

Google makes it easier to use Gemini without training AI on your data

AI training and chat history will now have separate toggles in the Gemini activity settings.

Read at the source Also covered by: TechCrunch

Open-source repositories 8 repositories

Trending on GitHub★ 13,148
Agents

backnotprop/plannotator

Plannotator is a local, browser-based review workspace that allows developers to inspect plans, code diffs, and artifacts created by AI coding agents. It connects directly via plugins and MCP, letting you annotate files and send feedback back to the agent without interrupting its session.

  • TypeScript
  • Apache-2.0
  • #agents
  • #code-review
  • #mcp
New on GitHub★ 5,824
Agents

KKKKhazix/agent-autoloop

Agent Autoloop is a lightweight automation system that runs command-line AI agents in the background based on scheduled rules or external events. It alerts you via desktop or chat apps only when a decision is needed, pausing without consuming system processes or tokens while waiting.

  • TypeScript
  • MIT
  • #agents
  • #automation
  • #cli
Open repository about +391 stars a day
Trending on GitHub★ 3,281
Agents

google-research/paper-orchestra

PaperOrchestra is a multi-agent framework that transforms raw research ideas and experimental logs into submission-ready LaTeX manuscripts. It orchestrates specialized agents to handle literature reviews, outline planning, section drafting, and visual chart generation.

  • Python
  • Apache-2.0
  • #agents
  • #research
  • #latex
Trending on GitHub★ 2,360
Agents

isair/jarvis

Jarvis is a local-first desktop AI assistant that supports natural voice and text interactions running on your own hardware. It can listen to discussions to jump in when addressed, securely managing your memory and personal data on-device with support for MCP tools.

  • python
  • #agents
  • #voice
  • #mcp
New on GitHub★ 2,017
Agents

mdiri-ai/SoduWork

SoduWork is a self-hosted engine designed for teams to collaborate with multiple AI agents across different models. It allows one agent to complete a task while another reviews it, with the request closing only after a human signs off.

  • JavaScript
  • MIT
  • #agents
  • #workflow
  • #self-hosted
Open repository about +100 stars a day
Hugging Face model♥ 220
Models

sensenova/SenseNova-Loop-9B

SenseNova-Loop-9B is a multimodal model designed to process both image and text inputs to generate text. It utilizes a looped-transformer architecture and is packaged using the safetensors format.

  • apache-2.0
  • #multimodal
  • #transformers
  • #vision
Hugging Face model♥ 211
Audio & video

meituan-longcat/LongCat-Video-Avatar-1.6

LongCat-Video-Avatar-1.6 is an open-source framework for generating audio-driven human avatar videos with coordinated facial and body motions. It supports stable long-form video generation exceeding 10 minutes while offering faster end-to-end inference and natural lip synchronization.

  • mit
  • #video
  • #avatar
  • #audio
Paper with code★ 27
Research

effl-lab/Weight-Space-Programming

This project investigates weight-space programming, where large language models directly write neural network weights without gradient-based training. It shows that LLM-designed sparse table and associative-memory architectures can beat same-sized Transformers on ARC tasks with far fewer parameters and faster GPU inference.

  • #research
  • #models
  • #transformers

Summaries are written automatically from the sources' headlines and excerpts and may contain mistakes. The full story is always at the source. To have a link or source removed, write to us from the contact section.

Let's start

Have an idea or a system that needs transforming?

Write us a few lines. Your message reaches us instantly, and we usually reply within one business day.

scouthub@lab: ~/contact

scouthub@lab:~$ contact --new --lang=en