AI agents are moving from controlled software environments toward a much wider question: how should they identify themselves on the open internet? Vint Cerf, one of the architects of the protocols behind the open internet, is now advising Innovation Labs as it works on that problem.
The organization is trying to create open architecture for AI agents to identify themselves. Its proposal, DNSid, would link agent identities to existing internet domain names and use cryptographic proofs to log registration over time.
Why agent identity is becoming urgent
Most AI agents today remain inside proprietary systems. They call on internal resources for defined purposes, rather than acting broadly across the open internet.
Businesses, however, are already imagining a more autonomous model. In that future, agents could operate across websites, services and organizations, and they could interact directly with other agents.
That raises a practical problem before it raises a futuristic one. If an agent takes an action online, other systems need some way to know what it is, where it came from, what authority it has and who is responsible for it.
Cerf described the issue to TechCrunch as a moment when naming and identification are becoming more important. He connected that shift to questions about authority, accountability, trust and where an agent’s identity is established.
What Innovation Labs is proposing
Innovation Labs is a subsidiary of Identity Digital, a DNS registry company. The company sees domain-name infrastructure as a practical foundation for agent accountability, especially if more online interaction begins to happen between agents rather than people.
The proposed system is called DNSid. Based on the source article, its purpose is to create identities for agents and connect each one to an existing internet domain name.
The system would also use cryptographic proofs to log an agent’s registration over time. That matters because an identity framework is not only about naming an agent once. It is also about creating a record that can be checked later.
Innovation Labs’ interim CEO Allie Kline says the company is trialing the standards with several unnamed hyperscalers and identity companies. The source does not identify those companies.
Why Cerf is involved
Cerf left Google after 20 years last week, according to the source article, but he is not stepping away from questions about the internet’s future. He is joining a handful of other internet luminaries who are lending their names to Innovation Labs’ effort.
His role is advisory, and the reason fits his long-running focus on internet architecture. The core question is not whether AI agents will be useful in isolation. It is whether they can work across the open internet in ways that other systems can recognize, trust and audit.
“I felt like I might be able to help them in a period of time when naming and identification is becoming increasingly important,” Cerf told TechCrunch.
He also warned that the work could be difficult because agents are more active than domains. A domain name may identify a place or organization online, but an agent may take actions. That difference makes accountability more complicated.
One open issue is what an organization is actually committing to when it registers an agent. The act of registration could imply responsibility, authority or trust, but the shape of that commitment is not yet clear from the emerging standards described in the source.
The standards problem
Innovation Labs is not the only group thinking about this. The source article says a variety of standards are beginning to emerge. That creates both momentum and risk.
If different companies use different agent systems that cannot work together, the internet could become fragmented at the agent layer. Cerf compared the pressure for interoperability to what happened with TCP/IP.
“Company X uses agent Y’s technology, and company A uses agent C’s technology, and then they don’t interwork with each other,” Cerf said.
For Cerf, functionality will be central to adoption. A standard is more likely to spread if it solves real problems for users and allows agents built with different technologies to interact.
That point is especially important because no agent can do everything people may want agents to do. If agents are going to become useful across the internet, they will need ways to coordinate beyond closed systems.
Trust, data and the agentic economy
Kline says one part of Innovation Labs’ proposal is that it does not include broader plans to do other kinds of AI business or own the registration data. In her view, a standard released by a hyperscaler and tied to proprietary data would face resistance.
That framing matters because identity systems can become powerful control points. If agent identity becomes a foundation for online interaction, the question of who operates it and who controls registration data becomes central.
Cerf is not treating an agent-driven internet as guaranteed. Asked whether the agentic economy is the internet’s destiny, he said he does not think it is inevitable.
But he does think people will try to build it. His reasoning is simple: if people can delegate tasks to an agent because it is easier, many will choose to do so.
That makes the identity question hard to avoid. Even if the agentic economy is not certain, the attempt to create it is already underway. Before AI agents can safely operate more freely across the open internet, the web may need a shared way to answer a basic question: which agent is this, and why should anyone trust it?