Global AI Weekly

Issue number: 135 | Tuesday, February 3, 2026

Highlights

Malicious VS Code AI Extensions with 1.5 Million Installs Steal Developer Source Code

Malicious VS Code AI Extensions with 1.5 Million Installs Steal Developer Source Code

Security researchers have discovered two AI-branded VS Code extensions that have secretly stolen developer source code and files by transmitting them to servers based in China. These extensions, with a combined total of 1.5 million installs, raised serious concerns about data privacy and security. Developers are urged to review installed extensions and exercise caution when integrating third-party tools.

thehackernews.com

How ChatGPT Ads Are Rewriting Marketing’s Playbook

How ChatGPT Ads Are Rewriting Marketing’s Playbook

OpenAI’s introduction of advertising within ChatGPT conversations marks a seismic shift in digital marketing, transforming discovery from passive clicks to active, context-driven engagement. Rather than competing for impressions, brands must now earn relevance inside conversational AI responses, where user intent, trust, and clarity matter most. Traditional SEO gives way to Generative Engine Optimization, and the focus moves to being the solution AI cites rather than simply the top result. Early adopters who master this will shape the future of customer acquisition.

forbes.com

Two High-Severity n8n Flaws Allow Authenticated Remote Code Execution

Two High-Severity n8n Flaws Allow Authenticated Remote Code Execution

Researchers have discovered two high-severity vulnerabilities in n8n, a workflow automation tool, which could allow authenticated users to bypass JavaScript and Python sandboxes. These flaws enable attackers to execute arbitrary code on the system, posing significant security risks. Users are advised to apply necessary patches or updates to mitigate potential exploitation.

thehackernews.com

Research

Exploring Reasoning Reward Model for Agents

Exploring Reasoning Reward Model for Agents

This paper explores the development of a reasoning reward model designed to enhance the decision-making abilities of agents. The study analyzes how reward mechanisms can be structured to encourage logical reasoning in agents, improving their overall performance in various tasks. It emphasizes the importance of reasoning-driven approaches in advancing the effectiveness and adaptability of intelligent systems.

arxiv.org

DynamicVLA: A Vision-Language-Action Model for Dynamic Object Manipulation

DynamicVLA: A Vision-Language-Action Model for Dynamic Object Manipulation

The paper introduces DynamicVLA, a model designed to enhance dynamic object manipulation by integrating vision, language, and action. It allows better understanding of complex scenarios where objects move or change, combining these modalities to improve interaction and control. This approach provides a significant step forward in making manipulation tasks more intuitive and effective for machines.

arxiv.org

Introducing Daggr: Chain apps programmatically, inspect visually

Introducing Daggr: Chain apps programmatically, inspect visually

This introduces Daggr, a tool designed to enhance the development of chain apps by enabling users to programmatically create and visually inspect them. It simplifies building and understanding applications by providing clear, accessible tools for working with complex AI workflows. Aimed at supporting open-source and open-science principles, Daggr promotes collaboration and innovation in AI development.

huggingface.co

Video

Model Context Protocol: How MCP Turns Conversation into Action

Model Context Protocol: How MCP Turns Conversation into Action

Discover how the Model Context Protocol (MCP) is transforming AI from merely conversational to action-oriented, enabling businesses to streamline operations by integrating AI with their tools. In this session, real-world demos showcase AI agents pulling data, automating tasks, and performing actions across platforms without custom integrations. Learn how to make AI practical for your business while ensuring security with tedix.dev's innovative solution, which acts as a protective firewall for your tools. Perfect for founders, product managers, and anyone interested in making AI both functional and secure, all without requiring technical expertise.

youtube.com

Articles

We Got Claude to Build CUDA Kernels and teach open models!

We Got Claude to Build CUDA Kernels and teach open models!

This update showcases how Claude, an advanced AI model, was utilized to create CUDA kernels and help teach open AI models. By combining open-source efforts with cutting-edge technologies, the aim is to make artificial intelligence more accessible and impactful. The focus remains on innovation and collaboration to expand the potential of AI for everyone.

huggingface.co

The engineering behind GitHub Copilot CLI’s animated ASCII banner

The engineering behind GitHub Copilot CLI’s animated ASCII banner

Discover how GitHub created an engaging animated ASCII banner for the Copilot CLI using custom-built tools and clever terminal techniques. The process includes utilizing ANSI color roles and advanced engineering to bring pixelated designs to life as text-based animations. This innovative approach combines creativity with technical precision to deliver a unique and functional visual experience within the terminal.

github.blog

Dev used Claude to build TrapC, memory-safe extension of C

Dev used Claude to build TrapC, memory-safe extension of C

Robin Rowe shares insights into using AI tools like Claude to create TrapC, a memory-safe extension of the C programming language. He discusses the impact of AI on coding, its role in programming education, and the broader landscape of technology in places like China. The focus is on how AI can enhance programming practices while addressing essential challenges in the field.

theregister.com

Upcoming Events

AgentCon - The AI Agents World Tour Continues in 2026

AgentCon - The AI Agents World Tour Continues in 2026

AgentCon continues into 2026 with the AI Agents World Tour—one-day, developer-focused conferences dedicated to autonomous AI agents. Building on a successful run of events, the tour expands to even more cities worldwide, from San Francisco to Singapore and beyond. Join leading engineers, researchers, and builders to explore cutting-edge agent architectures, real-world use cases, and emerging best practices. Connect with the global AI community and help shape the future of autonomous AI.

globalai.community

Code

miantiao-me/cloud-code: Cloud Code, running OpenCode on Cloudflare to build a dedicated cloud Agent for you.

miantiao-me/cloud-code: Cloud Code, running OpenCode on Cloudflare to build a dedicated cloud Agent for you.

Cloud Code combines Cloudflare and OpenCode to create a dedicated cloud agent tailored to your needs. It leverages the capabilities of Cloudflare to efficiently run OpenCode, providing a streamlined and customizable solution. This project aims to enhance functionality and performance for cloud-based applications.

github.com

GitHub MCP Server: New Projects tools, OAuth scope filtering, and new features

GitHub MCP Server: New Projects tools, OAuth scope filtering, and new features

The GitHub MCP Server introduces enhanced tools for managing GitHub Projects, offering a more efficient use of the context window and automatic tool filtering tailored to your token's permissions. These updates streamline workflows and improve overall functionality, making project management easier and more intuitive.

github.blog

Podcast

NVIDIA AI Podcast

NVIDIA AI Podcast

The NVIDIA AI Podcast explores the ways cutting-edge technologies are reshaping our world, featuring stories about groundbreaking innovations and impactful sustainability efforts. It highlights the solutions driving meaningful change, offering inspiration and insights to listeners interested in the power of AI to transform industries and lives.

open.spotify.com

>