Establishing AI Agent "Interconnection Standards," led by Google, OpenAI, and Anthropic

Global top technology giants are joining forces to promote the "infrastructure construction" in the field of artificial intelligence, aiming to break the interoperability barriers between AI Agents and enterprise applications by establishing unified technical standards. Competitors such as OpenAI, Anthropic, and Google have reached an agreement to jointly develop open-source standards to ensure that AI agents for automating white-collar tasks can work together stably and securely across platforms.

According to sources familiar with the project, the new alliance named "Agentic Artificial Intelligence Foundation" is expected to be announced as early as this week. In addition to the three core companies mentioned above, Microsoft, Amazon, and a number of other technology companies will also join. The foundation will be organized by the nonprofit organization Linux Foundation, which manages the Linux operating system, marking a rare collaborative model among competing tech giants at the underlying protocol level.

The core logic of this initiative is similar to the process of establishing interbank electronic payment standards in the global banking industry. By establishing universal connection rules, AI agents developed by different companies will be able to seamlessly integrate with various enterprise software, like trains running on a unified track, significantly reducing the technical barriers and fragmentation risks for enterprises deploying AI automation.

For the market, this signifies a key step for AI agents moving from a single tool to an interconnected ecosystem, which is expected to accelerate the commercialization process.

Despite the broad prospects for cooperation, the current technological environment still faces challenges. Chief Information Officers in enterprises generally report that while the relevant code writing is becoming increasingly common, the technology is not yet perfect, and all parties need to reach a consensus on processes such as security vulnerability remediation. The establishment of this standards organization is to formulate contribution rules for open-source software and share technological breakthroughs to address security and compatibility issues in practical applications.

Universal Standards Analogous to "Interbank Payments"

According to sources familiar with the matter, the main function of the Agentic Artificial Intelligence Foundation is to address the issue of AI agents being unable to effectively "communicate" in automated tasks. For AI agents to truly take over complex white-collar work, companies developing agents and those operating enterprise applications must reach an agreement on interconnected technical standards.

The operational model of this cross-industry organization will be led by the Linux Foundation. As a mature open-source project management organization, the Linux Foundation will coordinate the contribution rules for open-source code among the tech giants, avoiding ecological fragmentation caused by inconsistent standards.

This cooperation is seen as a positive signal for the industry's maturation. Although the tech industry has a long history of debate over open-source standards, in the emerging field of AI agents, the giants have chosen to cooperate to expand the market pie

Three Core Open Source Protocols

Insiders revealed to The Information that the organization will initially focus on standardizing three existing open-source AI tools, covering connection protocols, instruction formats, and local runtime agents:

First is the model context protocol invented by Anthropic. The MCP essentially serves as a way for AI models to communicate with application programming interfaces (APIs), aiming to standardize how AI agents connect with other applications.

Currently, the MCP has gained the most market traction, with companies including OpenAI, Microsoft, Google, and Cursor using it in their AI products or enterprise applications, such as ChatGPT and Google Workspace. Enterprises can connect ChatGPT to applications like Slack through the MCP, enabling it to read public channel conversations and answer management questions about customer account handling.

Second is Agents.md, invented by OpenAI. This is a format for issuing instructions to coding agents, standardizing the operational processes for agents to install applications or run tests on software.

Third is Goose, invented by Block. This is an open-source AI agent characterized by its ability to run locally on a single computer without relying on an internet connection.

Enterprise Implementation and Security Concerns

Despite the acceleration of the standardization process, enterprise-level applications still face severe security challenges. Alberto Martinez, IT director at AngelList, pointed out that as coding agents like Devin connect to operational tools like PagerDuty through the MCP to handle site outages, security has become a primary concern.

Martinez specifically mentioned the risk of "prompt injection attacks," where hackers infiltrate connected applications to trick AI agents into leaking customer data. He emphasized that while the MCP is convenient, it also brings numerous risks, as attackers can always find ways to deceive AI. To address this, AngelList has begun using tools from AI startup Runlayer to better track and manage applications connected through the MCP.

Although enterprise chief information officers indicate that writing code compatible with the MCP is rapidly becoming the norm for developing internal AI applications, there is a consensus that the existing technological framework still needs improvement. The newly established foundation will focus on addressing these specific security and technical coordination issues to pave the way for the large-scale commercialization of AI agents