互联网之父Vint Cerf，今年83岁了，还在给互联网打地基。
7月15日，TechCrunch报道称，Vint Cerf正在领导一个工作组，制定开放互联网上AI Agent的识别标准。简单说就是：给每个在公网上活动的AI Agent发一张「身份证」，让网站和服务能知道「正在访问我的是真人还是AI」。
这件事的分量，怎么强调都不过分。如果说MCP解决了「Agent怎么调用工具」的问题，A2A解决了「Agent之间怎么对话」的问题，那Vint Cerf要解决的就是更底层的问题——Agent在互联网上的身份问题。
为什么Agent需要「身份证」？
想象一下不久的未来：你公司的采购Agent在网上自动比价、下单；你的研究Agent在学术网站上自动读论文、做笔记；你的投资Agent在各个金融平台上自动操作。这些Agent每天要访问成千上万的网站，和无数的服务交互。
问题来了：网站怎么知道访问者是真人还是Agent？
现在的做法是验证码——reCAPTCHA、滑块验证、图片验证。但这些东西在AI面前越来越没用了——AI早就可以通过大多数验证码了。而且让Agent每次访问都去破解验证码，既浪费算力，又不优雅。
更深层的问题是责任。如果一个Agent在网上干了坏事——发垃圾邮件、刷流量、恶意爬取——谁来负责？是Agent的开发者？是运行Agent的用户？还是提供模型的厂商？如果Agent没有身份，责任就无法追溯。
Vint Cerf要做的，就是在互联网协议层面，给Agent一个可验证的身份标识。就像每个网站有SSL证书一样，每个Agent也可以有一个「Agent证书」——网站一看证书，就知道这个Agent是谁、属于谁、有什么权限。
这会是下一个TCP/IP吗？
说Vint Cerf是互联网的「总建筑师」一点都不夸张。上世纪70年代，他和Bob Kahn一起发明了TCP/IP协议——那是整个互联网的地基。没有TCP/IP，就没有今天的互联网。
现在，83岁的他又要给Agent时代打新的地基。如果这个标准真的做起来了，它的重要性可能不亚于当年的TCP/IP——因为Agent要在开放互联网上大规模活动，首先需要的就是「被识别」和「被信任」。
当然，这件事也有争议。比如：
隐私问题——如果每个Agent都有唯一身份，那用户的行为不就更容易被追踪了吗？Vint Cerf的思路是用零知识证明——网站可以验证「这是一个合法的Agent」，但不需要知道Agent具体是谁、属于谁。
去中心化问题——谁来发这些「Agent身份证」？如果是一家公司或者一个政府来发，那不就变成中心化控制了？目前的思路是类似SSL证书的体系——多个可信的第三方机构签发，互相认可。
执行问题——网站可以选择不验证Agent身份，或者假Agent可以伪造身份。标准是死的，人是活的，怎么保证大家都遵守？
这些问题都没有标准答案。但有一点是肯定的：Agent要真正走出封闭环境、走向开放互联网，身份问题是绕不过去的第一道坎。Vint Cerf提前看到了这个问题，并且开始动手解决——这就是互联网之父的远见。
明天见。
Vint Cerf, the father of the internet, is 83 years old and still laying foundations for the web.
On July 15, TechCrunch reported that Vint Cerf is leading a working group developing identification standards for AI agents on the open internet. Put simply: giving every AI agent active on the public web an "ID card" so websites and services can know "is this visitor human or AI?"
The significance of this can't be overstated. If MCP solves the problem of "how agents call tools" and A2A solves "how agents talk to each other," then what Vint Cerf is tackling is even more fundamental — the problem of agent identity on the internet.
Why Do Agents Need "ID Cards"?
Imagine the near future: your company's procurement agent automatically compares prices and places orders online; your research agent automatically reads papers and takes notes on academic sites; your investment agent autonomously operates across financial platforms. These agents visit thousands of websites every day and interact with countless services.
The question arises: how does a website know if a visitor is human or an agent?
Right now the solution is CAPTCHAs — reCAPTCHA, slider verification, image challenges. But these are becoming increasingly useless against AI — AI can already pass most CAPTCHAs. And making agents solve CAPTCHAs every time they visit something is both compute-wasteful and inelegant.
The deeper problem is accountability. If an agent does something bad online — sends spam, inflates traffic, maliciously scrapes content — who's responsible? The agent's developer? The user running the agent? The model provider? If agents have no identity, responsibility can't be traced.
What Vint Cerf wants to build is a verifiable identity layer for agents at the internet protocol level. Just like every website has an SSL certificate, every agent could have an "agent certificate" — a website checks the certificate and knows who this agent is, who it belongs to, and what permissions it has.
Will This Be the Next TCP/IP?
Calling Vint Cerf the "chief architect" of the internet is no exaggeration. In the 1970s, he and Bob Kahn invented the TCP/IP protocol — the foundation of the entire internet. Without TCP/IP, there would be no internet as we know it.
Now, at 83, he's laying new foundations for the agent era. If this standard really takes off, its importance could rival TCP/IP — because for agents to operate at scale on the open internet, the first thing they need is to be "identifiable" and "trusted."
Naturally, there are debates. For example:
Privacy concerns — if every agent has a unique identity, won't user behavior become easier to track? Vint Cerf's approach uses zero-knowledge proofs — websites can verify "this is a legitimate agent" without needing to know who the agent specifically is or who it belongs to.
Decentralization concerns — who issues these "agent ID cards"? If it's one company or one government, doesn't it become centralized control? The current thinking is an SSL certificate-like system — multiple trusted third-party issuers that recognize each other.
Enforcement concerns — websites can choose not to verify agent identities, or fake agents can forge identities. Standards are static, people are dynamic — how do you ensure everyone complies?
None of these questions have definitive answers. But one thing is certain: if agents are ever going to move beyond closed environments into the open internet, identity is the first hurdle they can't avoid. Vint Cerf saw this problem early and is already working on it — that's the foresight of the father of the internet.
See you tomorrow.
Agent要真正走出封闭环境、走向开放互联网，身份问题是绕不过去的第一道坎。
—— Dawn Vision编辑部
If agents are ever to leave closed environments for the open internet, identity is the first unavoidable hurdle.
— The Dawn Vision Editorial Desk
Vint Cerf · AI Agent · Internet Identity Standard · TCP/IP · Agent Protocol · Zero-Knowledge Proof
Sources · 信源 Sources
本文基于 Dawn Vision 认知引擎处理的 6 个源信号生成，经编辑部人工审核。素材来源：TechCrunch。
This article was generated by the Dawn Vision cognitive engine processing 6 source signals, with human editorial review. Source: TechCrunch.
