Agentic AI Foundation takes four agent projects from tech donors

The Agentic AI Foundation (AAIF) now manages four more open-source tools that help developers build AI agents. This brings its total to 10 projects. Alibaba, Anthropic, Hugging Face, and OpenAI gave control of OpenSandbox, Agent Skills, OpenEnv, and Open Responses to the foundation. The Linux Foundation created AAIF in December 2025 as an independent group that does not favor any one company.

The foundation started with three projects: Model Context Protocol (MCP), goose, and AGENTS.md. The four new projects help keep AI agents in safe, controlled spaces, give them instructions for tasks, create places where they can learn through practice, and make it easier for different large language models (LLMs) to work with the same systems.

OpenSandbox helps control what an AI agent is allowed to do. It uses tools for connecting software, systems that run programs in separate containers, controls for different stages of a program's operation, and parts that run inside the protected space. This means developers can limit what an AI agent can access without having to rebuild the AI model itself.

Agent Skills creates a standard way to write instructions that tell AI agents how to complete specific tasks. Anthropic said in a blog post that the project will remain open to everyone, guided by its community, and independent of any one company.

OpenEnv provides separate, container-based environments where AI agents can learn by trying things and getting feedback. These environments can work together and grow to support larger projects. Hugging Face gave OpenEnv to the foundation so teams can build places to test and train AI agents without having to set up complicated computer systems.

Open Responses, created by OpenAI, provides an open set of rules for how software communicates with large language models. These rules make requests, answers, live streams of information, and the ways AI models call other tools more consistent across different AI providers.

AAIF has grown to more than 300 members in less than a year. Sixty organizations joined during this quarter. New Gold-level members include Ant Group, Gates Foundation, MongoDB, Nutanix, Royal Bank of Canada, Vercel, and Vultr.

This growth shows that companies across the technology industry believe shared standards and tools for AI agents can help everyone, including competing companies. But the fast spread of AI agents also brings risks. According to Axios, banks in the United States are worried that AI agents are getting into customer accounts without permission from the banks, even when the customers themselves have agreed to let the agents access their accounts.

By managing these shared rules and tools, the foundation hopes to make AI agents work in more consistent and predictable ways when developers put them into use. But whether this works will depend on how developers use the tools and how quickly government regulators respond to AI agents accessing accounts without proper permission.