The co-creator of TCP/IP used his farewell address to warn that AI agents need formal protocols before the internet becomes an ungovernable mess of autonomous bots.
Vint Cerf, the man who co-designed the protocol suite that literally makes the internet work, is turning his attention to a problem that didn’t exist when he built TCP/IP in 1974: how do you identify and manage billions of AI agents operating autonomously online?
Cerf retired from Google on or around July 7, 2026, after 21 years as the company’s chief internet evangelist. In his final address at the Open Frontier conference, he didn’t talk about the good old days. He talked about the terrifying new ones.
The problem with letting AI agents wing it #
Cerf’s core argument is deceptively simple. Natural language, the thing humans use to communicate, is full of ambiguity. That’s fine when two people are sorting out dinner plans. It’s catastrophic when two AI agents are coordinating financial transactions at machine speed.
The need for precise, formal standards isn’t theoretical. It’s urgent. And Cerf would know. The RFC 675 specification he co-authored with Robert Kahn in December 1974 became the backbone of internet communication. TCP/IP was mandated as the US military protocol in 1982 and recognized as the internet standard on January 1, 1983.
The protocols already in play #
Cerf isn’t shouting into the void. Two major protocols have already emerged to address pieces of this puzzle.
Anthropic released its Model Context Protocol in November 2024. Think of MCP as a universal adapter that lets AI models plug into different data sources and tools without needing custom integrations for each one. Its software development kits have racked up 97 million monthly downloads.
Google’s Agent2Agent protocol, announced in April 2025, takes a complementary approach. A2A focuses on enabling different AI agents to communicate and collaborate with each other, regardless of which company built them. By April 2026, A2A had attracted over 150 organizational supporters.
Both protocols were donated to the Linux Foundation, a move that signals the industry wants these to be open standards rather than proprietary moats.
But neither MCP nor A2A fully solves the identity problem Cerf is highlighting. Interoperability is great. Knowing whether the agent you’re interacting with is legitimate, authorized, and accountable is a different challenge entirely.
Why crypto investors should be paying attention #
Look, there are no crypto tokens or blockchain projects directly connected to Cerf’s push for AI agent standards. Let’s be clear about that upfront.
The problem Cerf is describing, verifying the identity and authority of autonomous digital entities operating at scale, is essentially the same problem that decentralized identity protocols have been trying to solve for years. Projects building verifiable credentials, on-chain attestations, and decentralized identifier systems are working in the same conceptual space, just from a different starting point.
The risk is that the standards process moves entirely through traditional bodies like the IETF and W3C without meaningfully engaging blockchain infrastructure. TCP/IP itself was developed through government and academic channels. There’s no guarantee that AI agent identity standards will be any more crypto-native.
Investors watching this space should track whether the Linux Foundation’s stewardship of MCP and A2A begins to incorporate decentralized identity components, whether any of the 150-plus organizations supporting A2A include blockchain infrastructure providers, and whether Cerf himself, now freed from Google’s corporate constraints, begins advocating for decentralized approaches to the identity problem he’s outlined.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our