A Standard for AI Agent Identity Emerges — The Future of the Web Reshaped by AgentID

Until now, AI agents were nothing more than a nuisance for web services—'fake bots' that had to be blocked. However, the landscape is shifting dramatically, and agents are being reborn as 'first-class citizens' that can handle payments and collaboration with the safety of a verified ID, just like humans. We explore the fascinating evolution of an internet that is moving away from blind bot-blocking and toward embracing trusted, verifiable agents.

Digital Passports for AI: AgentID and AIT

Just as we show our passports at airport security to verify our identity, a movement has begun to provide AI agents with official credentials. This is the premise behind the 'AgentID' protocol currently under discussion by the IETF, and the 'Agent Identity Token' (AIT) utilized within it.

Much like a passport allows us to be verified as a legitimate traveler, these digital passports are presented when an agent accesses a website or platform. The token contains encrypted, transparent information regarding who the agent acts for, the actual owner, and the specific permissions they have been granted.

This allows websites to move past defensive, blanket-blocking postures. Instead, they can adopt a more flexible, intelligent approach: checking the encrypted ID and filtering for only verified, safe agents to grant them access.

Managing Tools Safely: The Intersection of COAZ and MCP

It would be incredibly convenient for an AI agent to read our emails and process payments based on our schedule. However, handing over full account permissions to an agent is akin to giving an unknown representative your personal seal. In the past, once an API key was shared, it was difficult to control exactly how the agent executed tools, leading to persistent security concerns regarding 'blind permission delegation'.

To neatly solve this, the OpenID Foundation approved the 'COAZ' profile. COAZ acts as an intelligent security guard that verifies in real-time whether a request is truly safe when an agent attempts to execute tools like the Model Context Protocol (MCP).

The secret lies in verifying the 'specific parameters' before the tool is even executed. Rather than granting broad permissions like 'allow email sending,' it applies granular filters, such as 'only allow if the recipient is a colleague and the body contains no sensitive information.' By using Common Expression Language (CEL), COAZ safely filters out dangerous commands or abnormal requests just before they run.

Thanks to this, both developers and users can confidently connect and use powerful agent tools without worrying about personal information or data leaks.

New Rules for Navigating the Web: agents-brief.txt

If the internet of the past relied on robots.txt to block indiscriminate scraping bots, we are now entering an era defined by peaceful coexistence and collaboration with agents. The recently refined agents-brief.txt (version 0.4.0) standard is a prime example. standard is a prime example.

Instead of blindly blocking agents, this configuration file allows website owners to declaratively state their preferred communication methods. It enables them to offer specific instructions to agents, such as, 'You may use our data for Retrieval-Augmented Generation (RAG), but use for AI model training is prohibited,' or 'To make a purchase or reservation on our site, please connect via the Model Context Protocol (MCP) endpoint.'

For instance, you can provide clear guidance at the website's root path like this:

# agents-brief.txt
Allow-Training: no
Allow-RAG: yes
Preferred-Execution: mcp

In fact, as hosting company DreamHost has begun deploying this configuration file as a default specification for domains within its service, an agent-friendly web ecosystem is rapidly taking hold. With agents moving about equipped with clear norms and identity authentication standards, the entire web ecosystem is now ready to welcome them as new digital guests.

From Blocking to Collaboration: How Platforms Welcome Agents

This trend is accelerating due to standardization efforts by the NIST and the emergence of commercial authentication solutions. For developers and platform operators, agents are no longer something to be reflexively blocked. It is time to treat them as new 'digital customers'—verifying their encrypted IDs and safely granting permissions so we can work together.