Agentic AI Foundation Adds 4 Projects, Expanding to 10 Total

The Agentic AI Foundation, a non-profit organization promoting the standardization of agentic AI, added four new projects under its umbrella at the 「AGNTCon + MCPCon North America 2026」held in North America in 2026. The four added projects are 「OpenSandbox」「Open Responses」「Agent Skills」and 「OpenEnv」, bringing the foundation's total project count to 10. This milestone has been achieved within one year of the foundation's establishment.

To understand this development, it is necessary to grasp the concept of 「MCP (Model Context Protocol)」. MCP refers to a common protocol through which AI agents—autonomous AI systems that perform tasks independently—can integrate with external tools and data. When different companies and developers each create their own integration methods, interoperability becomes difficult. This has led to increased industry-wide efforts to establish standardization. The Agentic AI Foundation is positioned as one of the organizations leading this standardization effort.

At this conference, Anthropic, OpenAI, Google, and UC Berkeley each delivered keynote speeches. The focus of these presentations centered on challenges in deploying MCP to actual products and business environments, security considerations, and design principles for cases where AI agents possess distinct identities.

The increasing emphasis on security reflects deepening challenges across the entire AI industry. As AI agents connect to external systems more frequently to retrieve data and execute tasks, the risks associated with communication pathways and access control management grow. The fact that security emerged as a major theme at this conference reflects the reality that the technology is transitioning from an 「experimental」phase to 「production deployment」.

The agenda of 「AI agents possessing distinct identities」is also an important future focal point. Clarifying which system an AI operates on and what authority it exercises when acting in place of humans is directly linked to corporate security policies and accountability. The fact that such design questions are now being addressed head-on at industry conferences signals the maturation of the field.

Detailed information about the functions and target domains of the four newly added projects remains limited at this time. However, project names such as OpenSandbox and OpenEnv suggest that preparing safe 「execution environments」for agents is recognized as a critical challenge. It can be seen that standardization and the construction of a secure operational foundation will be central axes for future discussions.

The fact that the Agentic AI Foundation has brought together 10 projects within one year of its establishment demonstrates strong industry interest in agentic AI standardization. The participation of major AI companies as keynote speakers further suggests that this movement extends beyond a niche developer community. How discussions on MCP security and operational design translate into concrete specifications and guidelines will be the next focal point of attention.