Vint Cerf backs push for internet standard to identify AI agents

Updated July 15, 2026 3:54 pm

Vint Cerf is helping launch a plan to give AI agents a verifiable identity on the open internet, a move that could shape how autonomous software is trusted, tracked and held accountable as it spreads beyond closed platforms. The internet pioneer has joined Innovation Labs just days after leaving Google, signaling that the next fight over the web may be about who, or what, is acting online.

The effort comes as companies race to build AI systems that can browse, buy, book, negotiate and coordinate with little human input. That future is still uneven and mostly confined to proprietary environments, but the need for a common way to identify these agents is becoming more urgent as more of the web begins to interact with software rather than people.

Why Vint Cerf is focusing on AI agent identity now

Cerf, one of the key designers of the internet’s foundational protocols, said the timing matters because the industry is beginning to confront a basic question: how do you know what an AI agent is, who created it, and what authority it has to act?

He described the problem as more than a technical formality. If an agent can make purchases, access records, trigger workflows or negotiate with another system, then identity becomes part of the trust model. Without a standard method for registration and verification, businesses may be forced to rely on fragmented private systems that do not interoperate well.

Cerf said the rise of AI agents has raised hard questions about who is accountable for an agent’s behavior, where its authority comes from and how its identity can be established in a way others can trust.

That framing reflects a larger shift in the AI market. Early consumer chatbots mainly answered questions. Newer agents are designed to take actions. That change pushes them closer to the core infrastructure of the web, where identity, authentication and audit trails matter just as much as speed or intelligence.

What Innovation Labs is proposing

Innovation Labs, a unit of Identity Digital, wants to create a shared system for identifying AI agents using internet domain infrastructure. Its proposal, known as DNSid, would tie an agent to an existing domain name and record its registration with cryptographic proofs over time.

The idea is to treat the domain namespace as a practical anchor for machine identity. In theory, that could make it easier for organizations, regulators and other software systems to determine whether an agent is authorized to act on behalf of a company, service or user.

Allie Kline, the interim chief executive of Innovation Labs, said the company is already testing the standards with several large cloud providers and identity firms, though she did not name them. The goal is to prove that the framework can work in the real world before any wider adoption attempt.

How DNSid would work

DNSid is intended to link an AI agent to a domain that already exists in the internet’s naming system. The registration would be recorded using cryptographic evidence, creating a traceable history of the agent’s identity over time.

That approach is designed to make impersonation harder and to provide a path for auditing. If an agent changes hands, changes duties or is retired, the record could help show when and how those changes occurred.

• Each agent would be connected to a domain name.
• Registration history would be protected with cryptographic proof.
• The system would be intended for verification and accountability.
• It would aim to work across organizations, not just within one vendor’s platform.

Why standards for AI agents are becoming a priority

The push for a standard reflects a broader concern that agentic systems are moving faster than the governance around them. Many current AI products work inside walled gardens, where the provider can control access, permissions and logging. But the next stage could involve agents crossing between services, vendors and organizations.

When that happens, a private internal policy is not enough. A business may want to know whether an incoming agent is genuinely authorized, whether it belongs to a known company, and whether its actions can be traced if something goes wrong.

That is especially important for enterprise workflows. An AI agent that can execute refunds, update records, communicate with suppliers or open support tickets needs more than intelligence; it needs a trustworthy identity layer that other systems can validate.

What problem is the industry trying to solve?

The industry is trying to solve the problem of machine-to-machine trust. As agents become more autonomous, the web needs a way to answer basic verification questions before one system allows another to act.

Without shared rules, every company may invent its own method for tracking agents, which would create friction and reduce interoperability. That would limit the usefulness of AI agents precisely at the moment when businesses want them to work across platforms.

Cerf sees adoption as a competition between competing systems

Cerf expects several standards to emerge, at least initially. He said the winner will likely be the protocol that proves most useful in practice, not simply the one with the most elegant design.

His view is shaped by the history of the internet itself. TCP/IP became dominant not because every company agreed on it in advance, but because users and developers pushed for a system that worked across networks and organizations. Cerf believes the same forces may shape agent identity.

Cerf argued that if one company’s agent technology does not interoperate with another’s, users will eventually pressure the market to settle on something broadly compatible, just as they did with TCP/IP.

That comment matters because it suggests the identity layer for AI may not be decided in a single committee room. Instead, it could be fought over by hyperscalers, identity vendors, standards groups and domain infrastructure companies, all trying to influence how the next generation of online activity is recognized.

How is this different from today’s internet identity systems?

It is different because AI agents are active participants, not just addresses or websites. A domain name identifies a location on the internet, but an agent can make decisions, issue requests and trigger real-world effects. That means identification is only part of the challenge; authorization and accountability also have to be defined.

In practical terms, an agent identity system must answer several questions that are far more complicated than those involved in registering a website:

1. Who created the agent?
2. What organization controls it?
3. What permissions has it been granted?
4. Can its history be audited later?
5. What happens when control changes?

Cerf said those issues are likely to be messy because no one has yet settled what it means to register an AI agent in a way that carries meaningful obligations. The challenge is not just to name the agent, but to define what that name signifies.

Why Identity Digital thinks domain infrastructure can help

Identity Digital is betting that the existing domain-name system offers a practical foundation because it is already a global internet utility. The company’s argument is that domain infrastructure is familiar, widely deployed and already tied to ownership and administration structures that can be mapped to digital accountability.

That matters in a future where AI agents may interact more with one another than humans do directly. If machine traffic grows, the internet may need a registry-like layer that is as ordinary and reliable as domain names are today.

Kline said Innovation Labs wants to avoid the perception that a major cloud or AI company could introduce a standard while also controlling the data behind it. She argued that organizations may resist a system if the same provider both sets the rules and captures the registration data.

Kline said there is likely to be resistance when a hyperscaler controls a standard and keeps the underlying information proprietary, because buyers will worry about data lock-in and lack of neutrality.

Why neutrality matters

Neutrality matters because standards only work at scale if competitors believe the rules are not rigged in favor of one platform. If a registry becomes a competitive moat, adoption could stall.

That tension is central to the broader AI infrastructure market. Cloud providers, model developers and identity firms all want influence over the systems that decide who or what gets access to digital services. A neutral registration layer could reduce friction, but only if the market trusts it.

Who else is likely to care about AI agent identity?

Almost every part of the digital economy could be affected, from cloud platforms and identity providers to enterprises, regulators and consumers. Any system that allows software to act on behalf of a person or business will eventually need proof that the software is legitimate.

That includes sectors such as finance, retail, customer service, logistics and digital marketing, where automated agents may eventually handle transactions or coordinate with external services. The more autonomy agents gain, the more valuable a reliable identity layer becomes.

The issue is also likely to draw attention from policymakers. If AI agents are capable of making decisions that affect consumers or markets, then identifying them may become part of broader governance and security debates. A registry alone will not solve those questions, but it could become a building block for enforcement.

What Cerf thinks the future really looks like

Cerf does not present an all-or-nothing vision of an “agentic internet.” He said it is not guaranteed that the web will evolve in that direction. But he also believes people will absolutely try to use agents whenever they can, because delegating tasks to software is convenient and efficient.

That realism is important. The future he describes is not driven by ideology but by behavior. If users discover that agents can save time, they will choose them. If companies find agents can reduce labor costs or speed up workflows, they will deploy them. The question is not whether the trend exists, but what guardrails will accompany it.

Cerf said people are likely to adopt agents whenever they can because humans are inclined to choose the easier option when software can do the work for them.

That statement captures both the promise and the risk. Convenience drives adoption, but convenience without identity, oversight and auditability can lead to abuse. The more powerful the agent, the more important the infrastructure behind it becomes.

What happens next?

For now, Innovation Labs is still in the early stages of trying to make its approach a practical standard. Its talks with major technology and identity companies suggest momentum, but not consensus. Cerf’s involvement may help bring additional attention, yet the hard work will be persuading the market that this is the right way to handle agent identity.

The outcome will depend on whether the proposed system can balance openness, trust and usability. If it is too closed, companies may reject it. If it is too loose, it may fail to provide the accountability it promises. If it works, it could become one of the quiet but essential pieces of infrastructure that make the agent economy possible.

That is why the project is drawing attention beyond the niche of internet standards. It is about deciding how the next wave of online action will be labeled, verified and governed. And for Cerf, who helped define the architecture of the modern internet, that is a familiar kind of frontier.