Vint Cerf just joined a startup trying to give AI agents a verifiable ID on the internet
One of the people who built the internet thinks AI agents need passports. He's now helping design them.
Key points
- Vint Cerf, co-creator of the TCP/IP protocols that underpin the modern internet, left Google in May 2025 after 20 years and immediately joined Innovation Labs as an adviser.
- Innovation Labs, a subsidiary of DNS registry company Identity Digital, has proposed a system called DNSid to give AI agents verifiable, trackable identities online.
- DNSid links each AI agent to an existing internet domain name and uses cryptographic proofs, tamper-resistant digital signatures, to log registrations over time.
- Innovation Labs' interim CEO Allie Kline says the system is already being trialled with several large cloud and identity companies, though she has not named them.
- No single identity standard for AI agents exists yet, and multiple competing proposals are in play.
Vint Cerf helped invent the rules that let computers talk to each other across the early internet. Technologists notice when he moves. So when he walked out of Google after two decades and signed on to advise a startup, the question worth asking is: why this, why now.
Cerf told TechCrunch it comes down to accountability. Who is responsible when an AI agent does something wrong?
An AI agent is software that can carry out multi-step tasks on its own, booking a flight, placing an order, sending emails, without a human clicking each step. Most agents today work inside a single company's closed system. Businesses are already building agents that will roam the open internet, talk to other agents, and take real actions with real consequences.
The problem is obvious. If an agent books a fraudulent hotel room or leaks your data, who owns that mistake? Right now there is no agreed way to check an agent's identity, trace who authorised it, or hold anyone accountable.
Innovation Labs wants to fix that with DNSid. Every AI agent gets a registered identity tied to an existing internet domain name, the kind that ends in .com or .org. A cryptographic proof, a mathematical stamp that cannot be quietly altered later, logs that registration permanently. Think of it as a birth certificate for software.
Innovation Labs sits inside Identity Digital, which already runs domain-name registries, the databases that match website addresses to servers. Building agent IDs on top of that existing infrastructure avoids creating yet another parallel network from scratch. We looked at how AI agents are already tripping up enterprise workflows in our 14 July report on why confident wrong answers keep happening, and the absence of any accountability layer was a thread running through that whole story.
Cerf draws a direct parallel to how TCP/IP, the communication rules he co-wrote in the 1970s, became the universal standard. Not because a committee voted on it, but because users demanded that systems work together. "Nobody can do everything that you might want every agent to do," he told TechCrunch. The same user pressure, he believes, will eventually force rival agent systems to interoperate.
One detail that matters for anyone worried about big-tech control: Innovation Labs says it will not use registration data for other AI business and will not own the data itself. Kline specifically flagged the risk of a large cloud company releasing a standard and quietly hoarding the data that flows through it.
Cerf is candid that none of this is guaranteed. "I don't think it's inevitable," he said of an agent-driven internet. But he added: "We are fundamentally lazy creatures, and if we find a way to have an agent do something for us, we're very likely to choose to do that."
Should ordinary people care about AI agent identity standards?
Yes. The answer to "who is responsible" determines whether you can get a refund or your data corrected. There is no reliable way right now to audit which company sent the agent that acted on your behalf or against your interests. A working identity standard changes that.
My read: the architecture here is genuinely sensible, domain names are already universal, and Cerf's TCP/IP analogy is not just flattering spin. What to watch is whether the unnamed hyperscalers trialling DNSid adopt it publicly or quietly build their own walled version instead. That is the real test, and it will tell you whether this becomes infrastructure or just a press release.
