Vint Cerf Is Building a Standard to Identify AI Agents on the Web
Internet pioneer Vint Cerf is developing a proposed standard for identifying and authenticating AI agents operating on the open internet. The effort aims to bring accountability and interoperability to autonomous agents before they become ungovernable infrastructure.
Original sourceVint Cerf, one of the co-inventors of TCP/IP and a person who has some credibility when it comes to protocols that actually stick, is working on a framework for giving AI agents verifiable identities on the open internet. The core problem he's addressing: as AI agents proliferate and start acting on behalf of users — browsing, transacting, communicating — there's currently no standard way to know who or what you're talking to, who is responsible for it, or whether it has permission to do what it's doing.
The proposal would establish a layer for agent identification — think something analogous to TLS certificates or WHOIS, but for autonomous software actors. An agent acting on your behalf to book a flight or send an email would carry a verifiable credential tying it to an identity, an owner, and presumably some scope of authorization. Without something like this, the internet increasingly becomes a space where humans can't distinguish other humans, human-operated software, or fully autonomous agents.
This isn't a finished spec — it's a working effort, and the details of governance, adoption path, and enforcement are still open questions. But Cerf's involvement matters because he has both the institutional credibility to push something through standards bodies and a long track record of thinking about what makes internet infrastructure actually durable. The difference between a good idea and a standard is usually not the idea.
The stakes here are significant. Without an identification layer for agents, the incentives push toward opacity: companies benefit from agents that are hard to block, scrape, or rate-limit. A working standard would shift that calculus, giving websites, services, and regulators a handle on an increasingly autonomous web. Whether the industry coalesces around a single approach — or fragments into competing identity schemes the way it has with other attempted internet standards — remains the real open question.
Panel Takes
The Builder
Developer Perspective
“The primitive here is basically a credential format plus a resolution mechanism — something like a DID or an SPIFFE SVID but scoped to agent authorization contexts. The interesting DX bet is whether this ends up as a header, a well-known endpoint, or something you have to integrate into your auth stack from scratch. I'd be watching the first reference implementation closely: if it requires a new server-side library before you can do anything, adoption will crater regardless of how sound the spec is.”
The Skeptic
Reality Check
“The graveyard of internet standards that made complete sense and went nowhere is long — OpenID, P3P, POWDER — and agent identity has the same coordination problem every one of them had: the parties who most need to comply have the least incentive to do so. Cerf's credibility is real, but credibility didn't save DNSSEC from a decade of partial deployment. What kills this in 12 months isn't a better competing proposal — it's that Google, Anthropic, and OpenAI each ship their own agent identity schemes and call it a day.”
The Futurist
Big Picture
“The thesis here is falsifiable and important: within three years, a meaningful fraction of internet traffic will be agent-to-service rather than human-to-service, and the existing trust infrastructure — built for browsers and humans — will be structurally inadequate. If that's true, whoever defines the identity layer defines the permission model for autonomous action on the web, which is a massive second-order power shift toward whoever controls credential issuance. The trend Cerf is riding is real and he's early, not on-time — which means the standard has a chance to actually shape behavior rather than ratify it.”
The Founder
Business & Market
“There's no direct business model here, but there's a platform play hiding inside it: whoever builds the canonical agent identity registry or the tooling layer on top of this standard is in a structurally strong position. The moat question is whether this becomes open infrastructure like certificate authorities — technically open, practically oligopolistic — or stays genuinely distributed. Either way, the companies betting on agents as a primary interface should be funding this work, because the alternative is building on a foundation where any regulator or platform can pull the rug by demanding identity verification they can't provide.”