Tools

Vint Cerf joins effort to build an open naming architecture for AI agents

Vint Cerf, a co‑author of the internet protocols, has left Google after 20 years and begun advising Innovation Labs on standards for identifying AI agents.

Vint Cerf joins effort to build an open naming architecture for AI agents

Vint Cerf, one of the architects of the protocols underpinning the open internet, left Google last week after twenty years and has begun advising Innovation Labs. The organization is working on an open architecture to enable AI agents to identify themselves online.

What Innovation Labs proposes

Innovation Labs is a subsidiary of Identity Digital, a company that operates DNS registry services. Identity Digital views domain‑name infrastructure as a pragmatic foundation for making AI agents accountable and for positioning itself for a future in which more online interactions occur between agents than between humans.

Innovation Labs has proposed DNSid, a registry for agent identification that links each agent to an existing internet domain name and uses cryptographic proofs to log its registration over time. Allie Kline, Innovation Labs’ interim CEO, said the company is trialing the standards with several unnamed hyperscalers and identity firms.

Why a shared standard is needed

Today most AI agents operate inside proprietary systems and call internal resources for specific tasks. Businesses, however, are already imagining agents acting with greater autonomy across the internet and interacting directly with one another. A major roadblock so far has been the absence of a shared standard for identifying and auditing agents.

Cerf highlighted the novel questions this raises: what authority does an agent have, where did that authority derive from, who is accountable for the agent’s behavior, and how and where is the agent’s identity established and proven? He warned these issues will be thorny because agents will be far more active than domain names, and it is not yet clear what commitments an organization makes when it registers an agent.

Interoperability and adoption

Cerf noted multiple technical approaches are under consideration and argued that widespread adoption will depend on practical functionality. He cited the historical example of TCP/IP, where pressure from users helped consolidate standards. If different companies deploy incompatible agent technologies, user demand will likely drive convergence toward interoperable solutions.

Kline stressed that Innovation Labs’ proposal does not come bundled with broader AI business ambitions nor with an intent to own registration data. She suggested there would be resistance if a hyperscaler published a standard while retaining proprietary data tied to that standard.

Outlook

Cerf said he does not consider an agent‑driven internet inevitable, but expects many will attempt to automate tasks through agents because people prefer easier solutions. Going forward, the field will need to resolve technical identification methods, accountability frameworks, and interoperability — developments that will shape the evolution of the internet and the services built on it.