Vint Cerf Wants to Give AI Agents an ID Card for the Open Internet

The headline in one line

Vint Cerf — the guy whose name is on the protocols that made the internet an internet — is now advising Innovation Labs on a proposal called DNSid: a way to give every AI agent a verifiable identity anchored to a real internet domain. Reported first by TechCrunch on July 15, 2026.

Translation for builders: the plumbing for agent-to-agent commerce on the open web is being drafted right now, and it is not going to be owned by one hyperscaler if Cerf gets his way.

Why this matters more than another spec announcement

Most AI agents today live inside a single vendor''s walled garden. Your ChatGPT agent calls ChatGPT tools. Your Claude agent calls Claude tools. They mostly don''t meet strangers on the open internet — and when they do, there is no shared way to answer three questions every counterparty needs to answer:

1. Who is this agent?

2. Who is accountable for what it does?

3. Should I trust its cryptographic proof of that identity?

DNS solved that problem for websites in the 1980s. DNSid is trying to solve it for agents in the 2020s — by linking each agent identity to an existing domain name and logging cryptographic proofs of registration over time. That gives every agent a paper trail that traces back to a real registered entity, the same way `stripe.com` traces back to Stripe Inc.

What DNSid actually proposes (in plain English)

Innovation Labs, a subsidiary of DNS registry company Identity Digital, has submitted a draft standard that does three things:

Interim CEO Allie Kline told TechCrunch the company is already trialing the standard with "several unnamed hyperscalers and identity companies." Cerf''s job is not to write the spec — it is to give it interoperability credibility in a moment when three or four competing standards are all racing to become the default.

Why Cerf is the right name attached to this

"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 told TechCrunch. "Nobody can do everything that you might want every agent to do… and so we''re going to have to rely on the pressure coming from the users. This is what happened with TCP/IP."

That is the whole quote you should tattoo on the wall. Interop wins because users demand it, not because a vendor decrees it. The reason the open internet exists at all is that Cerf and a handful of engineers refused to let it be a private network of AOL-like walled gardens. He is essentially arguing the same fight is coming for the agent economy, and DNSid is one attempt to keep the plumbing open.

The competing world where this could fail

Cerf himself is clear-eyed about the risk. Multiple standards are emerging — Google''s Agent2Agent (A2A), Anthropic''s MCP on the tools side, various W3C proposals, and now DNSid. If the industry fragments, agents from different vendors won''t interoperate, and the "agentic internet" will be a patchwork of proprietary silos that look a lot like today''s enterprise SaaS map.

Kline''s pitch is that DNSid has one structural advantage: Innovation Labs doesn''t want to also sell you the AI. She told TechCrunch, "there''s a lot of organ rejection to a hyperscaler releasing [a standard] and having that proprietary data." Neutral registry + neutral standard is the same playbook that made DNS itself work.

The 88 Labs take — what builders should actually do this quarter

Standards fights take years. But the implications for anyone shipping agent products are actionable now. Here is our short list.

1. Assume your agents will need verifiable identities within 18 months

Enterprises are already asking "who is this bot talking to my API and who do I sue if it goes rogue?" If you sell agent products into regulated buyers — finance, healthcare, legal, government — an answer to that question is going to move from "nice to have" to a procurement gate. Design your agent architecture so identity is a first-class field, not a header you add later.

2. Own the domain your agents will speak from

If DNSid or something like it wins, your agent identity is going to be rooted in a domain you control. Buy the domains you''ll want for agent surfaces now (`agent.yourcompany.com`, `bookings.yourcompany.com`, etc.) and set up delegation properly. Domain squatting on agent-friendly subdomains is going to be a thing.

3. Log everything an agent does with a durable identifier

Whatever standard wins, the audit story is the same: prove which agent did what, when, on whose authority. Start logging structured events with a stable agent identifier per action today. When the standard lands, you migrate the identifier field — you don''t rebuild your telemetry.

4. Do not bet the company on any single standards horse

Cerf''s own quote — "nobody can do everything that you might want every agent to do" — is code for there will be bridges. Build an abstraction layer so your agents can present identity in whatever format a given counterparty expects (DNSid today, A2A tomorrow, MCP-flavored the day after).

5. Treat "an agent did it" as a legal category, not a technical one

The interesting quote from Cerf is not about the protocol. It is this: "what authorities they have, where they have derived those authorities, who is accountable for the behavior of an agent." Your terms of service, your customer contracts, and your incident response runbooks all need to answer those questions before your agents start transacting with strangers.

The bigger picture

The 1980s internet worked because a small group of neutral people insisted that no single company should own the address book. That fight is not over — it is happening again, at a higher layer of the stack, with agents instead of hostnames.

You can be cynical about whether DNSid specifically wins. You should not be cynical about whether something like it is coming. The demand for accountable, interoperable, cryptographically verifiable agent identity is real, and it will be filled — either by an open standard or by whichever hyperscaler moves fastest to make theirs the default.

Cerf''s bet is that the open standard wins if enough builders show up early. That is a bet worth taking.

FAQ

What is DNSid? DNSid is a proposed standard from Innovation Labs (a subsidiary of Identity Digital) that assigns AI agents cryptographically verifiable identities anchored to real internet domain names, so agents can prove who they are and who is accountable for their actions.

Why is Vint Cerf involved? Cerf co-designed TCP/IP, the protocols that made the open internet possible. He is advising Innovation Labs to lend the standard interoperability credibility during a period when multiple competing agent-identity standards are emerging.

How is DNSid different from Google''s A2A or Anthropic''s MCP? A2A and MCP focus on how agents talk to other agents and tools. DNSid focuses one layer down — proving which agent you are actually talking to and tying that identity to an accountable registered domain owner.

Should I wait for a winning standard before building? No. Design your agents so identity is a first-class, swappable field. Log every agent action with a durable identifier now. When a standard consolidates, you migrate the field — you don''t rebuild your product.

Is DNSid production-ready today? No. Innovation Labs is trialing it with hyperscalers and identity companies. The IETF draft is early. Treat it as a signal about where enterprise procurement questions are heading, not as a spec to implement in production this quarter.

Where can I read the original TechCrunch story? Tim Fernholz''s report is at techcrunch.com/2026/07/15/vint-cerf-is-working-on-a-plan-to-unleash-ai-agents-on-the-open-internet.

Related reading