🔶 Agent Name Service：通用 AI 智能体身份系统

有个笑话说，如果互联网出了问题，罪魁祸首总是域名系统（DNS）。这是因为……嗯，实际上经常确实是 DNS 的锅。话虽如此，我们确实需要一个通用、开放的方法来追踪 AI 智能体，那为什么不试试 DNS，看看互联网的通用目录能否不仅标识 IP 地址，也能标识智能体呢？

Linux 基金会正在尝试这个想法，推出了 Agent Name Service（ANS）。这是一个开放、受 DNS 启发的标准，旨在为 AI 智能体提供可验证的、互联网原生的身份，以及一种安全的发现与信任方式。

为什么是 DNS？因为尽管 DNS 出现故障时后果严重，但它其实并不经常出问题。毕竟，DNS 每秒在全球处理超过 1 亿次查询。

因此，与其引入新的查找网络或专有生态，ANS 选择将智能体身份直接锚定到 DNS。这个提议的框架使系统和用户能够验证智能体代表谁、拥有什么权限，以及其代码和运行历史是否保持真实且未被篡改。

这是一项至关重要的工作。正如 Linux 基金会执行董事 Jim Zemlin 所说：“AI 智能体将越来越多地跨企业、平台和数字服务运行，这使得可信的身份基础设施成为一项基本要求。通过基于 DNS 和开放标准构建，ANS 为全球数字经济中经过验证的智能体通信创建了一个可扩展且可互操作的框架。”

这样一个系统之所以必要，是因为即使企业迅速接纳了自主智能体，它们也缺乏一种通用的、供应商中立的方式，来了解到底谁真正控制着某个智能体、它运行着哪个代码和模型版本，以及它的元数据或端点是否被篡改过。来自云和 AI 供应商的专有“智能体注册表”以孤岛的方式解决这个问题。ANS 旨在成为跨越这些孤岛的粘合剂，甚至可能取代它们。

一个 ANS 名称会锚定到一个域名，因此由某家公司运营的智能体会明确地与该公司的互联网身份绑定。解析过程对任何熟悉 DNS 的人来说都很熟悉，但增加了安全层。当一个应用或另一个智能体请求 ANS 解析一个名称时，服务会返回经过签名的记录，这些记录可以在任何交互发生前，根据公钥基础设施和证书进行校验。其理念是，系统在信任一个智能体之前，可以程序化地回答三个问题：谁运行它、它是什么版本，以及记录自注册以来是否被更改过。

支持者认为，这种方法赋予了智能体在 Web 身份结构中的“一等公民”地位，而不是把它们当作隐藏在 API 背后的不透明服务。然而，也有其他人担心，由于智能体 AI 需要不断调用和召回智能体来完成工作，ANS 可能会太慢。

ANS 规范并非凭空出现。这项工作源自 OWASP GenAI 安全项目。贡献者们一直在苦苦思索，如何在不将企业置于专有注册表和模糊发现机制之下的情况下，确保多智能体系统的安全。

在 OWASP 内部，ANS 被定位为一个与协议无关的智能体发现和注册层，可以与现有的 DNS 基础设施并存。其设计依赖于公钥基础设施和签名的注册记录，并考虑集成去中心化标识符（DID）和法人机构识别编码（LEI），以便智能体可以通过加密方式关联到真实的组织。这种关联意在解决“影子 AI”问题，即智能体在没有明确出处或治理的情况下被创建出来，却可能做出影响安全的关键决策或访问敏感数据。

为了将这个想法转变为可部署的标准，两家公司，Infoblox 和 GoDaddy，一直在努力打造具体的服务。

首先，Infoblox 正在推进用于 AI 发现的 DNS（DNS-AID）。这是一种基于 DNS 构建的、开放且可互操作的智能体发现方法。同时，GoDaddy 正在帮助开发 ANS。这两项工作是互补的，并且正在社区标准机构中开发，其明确目标是实现独立实施，避免单一或集中的供应商控制。

通过将 ANS 作为自己的倡议来托管，该基金会表明智能体身份和发现正成为一个跨行业的关注点，涉及云提供商、企业软件供应商、注册商和安全团队。他们还希望，Linux 基金会的这一举措能有效地将 ANS 从一个以安全为主的项目升级为一项更广泛的标准努力。

在仍在成形中的计划里，DNS-AID 定义了一种命名约定和元数据格式，这些格式分层建立在现有 DNS 记录类型之上，例如 SVCB、TXT 和 TLSA。这一切都由 DNSSEC 来确保安全。在实践中，组织可以以类似 _chatbot._ mcp._agents.example.com 这样的名称发布智能体，用 DNSSEC 对该区域进行签名，并让其他智能体在通过 MCP、A2A、HTTPS 或任何可通过 DNS 的服务绑定（SVCB） 和应用层协议协商（ALPN）来寻址的协议进行连接之前，解析并验证这些记录。

许多公司都在支持 ANS。Cloudflare、Equinix、GoDaddy、ISC 以及其他基础设施提供商已作为早期支持者加入。他们希望 AI 智能体发现能存在于开放的互联网标准中，而不是被锁定在少数平台提供商手中。Linux 基金会则将其定位为让 ANS 和 DNS-AID 成为一个更广泛的“智能体 AI”基础中的构建块，这个基础最终可能会与 Kubernetes 及其他旗舰项目一起进入其项目组合。

他们的动机很直接。如果 ANS 作为一个中立的命名和身份层获得牵引力，他们就可以将智能体插入到现有的域、DNS 和证书生态系统中，而不是为 AI 重新发明轮子。

注册商和 DNS 提供商看到了一个可供管理的新名称和记录类别。安全厂商看到了一个标准化的地方，可以在其上附加策略和监控。云提供商则看到了一个无需放弃对其自身智能体编排栈的控制，就能实现互操作的机会。

并非所有人都对这个提议感到满意。一个开放的、锚定于 DNS 的智能体身份标准，会将部分控制权从垂直整合的 AI 平台手中移走，这些平台一直在构建自己的发现和信任机制。如果 ANS 成功，企业可能会坚持要求任何供应商的智能体都必须在其自己的域名下注册并可验证，并进行独立的身份校验。这可能会让那些依赖封闭智能体目录的专有生态系统变得复杂。

ANS 与一个相关的工作同时到来，那就是 Linux 基金会的用于 AI 发现的 DNS（DNS-AID）。DNS-AID 专注于另一个互补的问题：帮助 AI 智能体和模型上下文协议（MCP）服务器相互发现和验证。顾名思义，DNS-AID 也使用 DNS 作为一个中立目录。

就目前而言，ANS 更像是一张路线图而非现实。但 Linux 基金会决定接过这项工作，表明智能体身份和发现即将成为一等一的标准问题，而不仅仅是埋藏在专有 AI 仪表盘中的功能。随着自主智能体的激增，“我正在交谈的这个东西到底是谁？”这个问题，可能会变得与驱动它们的模型一样重要——而 ANS 正是试图用互联网已经理解的语言来回答它：名称、密钥和域名。

In partnership with

The joke goes that if something goes wrong with the Internet, it's always Domain Name Server (DNS). That's because, uh, well, actually it often is DNS. That said, we do need a universal, open way to track AI Agents, so why not try DNS to see whether the Internet's universal directory can identify Agents as well as IP addresses?

The Linux Foundation is trying this idea by introducing the Agent Name Service (ANS). This is an open, DNS‑inspired standard designed to give AI agents verifiable, internet‑native identities and a secure way to be discovered and trusted.

Why DNS? Well, even though DNS breaks down badly when it fails, it doesn't really fall down that often. After all, DNS processes over 100 million queries per second worldwide.

So, rather than introducing a new lookup network or proprietary ecosystem, ANS anchors agent identity directly to DNS. This proposed framework enables systems and users to verify who an agent represents, what permissions it has, and whether its code and operational history remain authentic and unchanged.

This is an essential effort. As Jim Zemlin, CEO of Linux Foundation, said, “AI agents will increasingly operate across enterprises, platforms, and digital services, which makes trusted identity infrastructure a foundational requirement. By building on DNS and open standards, ANS creates a scalable and interoperable framework for verified agent communication across the global digital economy.”

Such a system is necessary because even as enterprises rapidly embrace autonomous agents, they lack a common, vendor‑neutral way to know who really controls an agent, what code and model version it is running, and whether its metadata or endpoint has been tampered with. Proprietary “agent registries” from cloud and AI vendors address the problem in silos. ANS is intended to be the glue that spans them, or possibly replaces them.

An ANS name is anchored to a domain, so an agent operated by a company is explicitly tied to that company’s Internet identity. The resolution process looks familiar to anyone who has worked with DNS, but with added security layers. When an application or another agent asks ANS to resolve a name, the service returns signed records that can be checked against public key infrastructure and certificates before any interaction occurs. The idea is that a system can programmatically answer three questions before trusting an agent: who runs it, what version it is, and whether the record has been altered since it was registered.

Supporters argue that this approach gives agents “first‑class status” in the web’s identity fabric, rather than treating them as opaque services hidden behind APIs. Others, however, worry that ANS will be too slow since agentic AI is constantly calling and recalling agents to get work done.

The ANS specification did not appear out of thin air. The work emerged from the OWASP GenAI Security Project. These contributors were grappling with how to secure multi‑agent systems without leaving enterprises at the mercy of proprietary registries and opaque discovery mechanisms.

Within OWASP, ANS has been framed as a protocol‑agnostic agent discovery and registration layer that can sit alongside existing DNS infrastructure. The design leans on public key infrastructure and signed registration records, and it contemplates integrating decentralized identifiers (DIDs) and legal entity identifiers (LEIs) so that agents can be cryptographically tied to real organizations. That linkage is intended to address the “shadow AI” problem, in which agents are spun up with no clear provenance or governance, yet can make security‑critical decisions or access sensitive data.

To transform this idea into a deployable standard, two companies, Infoblox and GoDaddy, have been working on making concrete services.

First, Infoblox is advancing DNS for AI Discovery (DNS-AID). This is an open, interoperable approach for agent discovery built on DNS. Simultaneously, GoDaddy is helping develop ANS. Both efforts are complementary and are being developed in community standards bodies with the explicit goal of enabling independent implementations and avoiding single- or concentrated-vendor control.

By hosting ANS as its own initiative, the Foundation is signaling that agent identity and discovery is becoming a cross‑industry concern that touches cloud providers, enterprise software vendors, registrars, and security teams alike. They're also hoping that

The Linux Foundation’s move effectively graduates ANS from a security‑focused project into a broader standards effort.

In the plan, which is still coming together, DNS-AID defines a naming convention and metadata format layered on top of existing DNS record types such as SVCB, TXT, and TLSA. This is all secured by DNSSEC. In practice, an organization can publish agents under names like chatbot. mcp._agents.example.com, sign the zone with DNSSEC, and let other agents resolve and verify those records before connecting over MCP, A2A, HTTPS, or any protocol addressable via DNS's Service Binding (SVCB) and Application-Layer Protocol Negotiation (ALPN).

Many companies are supporting ANS. Cloudflare, Equinix, GoDaddy, ISC, and other infrastructure providers have signed on as early backers. They want AI agent discovery to live in open Internet standards rather than be locked into a handful of platform providers. The Linux Foundation, for its part, is positioning ANS and DNS-AID as building blocks in a broader “agentic AI” foundation that could eventually sit alongside Kubernetes and other flagship projects in its portfolio.

Their motivation is straightforward. If ANS gains traction as a neutral naming and identity layer, they can plug agents into existing domain, DNS, and certificate ecosystems rather than reinventing the wheel for AI.

Registrars and DNS providers see a new category of names and records to manage. Security vendors see a standardized place to attach policy and monitoring. Cloud providers see an opportunity to interoperate without ceding control of their own agent orchestration stacks.

Not everyone is happy with this proposal. An open, DNS‑anchored standard for agent identity shifts some control away from vertically integrated AI platforms, which have been building their own discovery and trust mechanisms. If ANS succeeds, enterprises may insist that agents from any vendor be registered and verifiable under their own domains, with independent identity checks. That could complicate proprietary ecosystems that rely on closed agent directories.

ANS arrives alongside a related effort, the Linux Foundation's DNS for AI Discovery (DNS-AID). DNS-AID focuses on the complementary problem of helping AI agents and Model Context Protocol (MCP) servers find and verify one another. As the name says, DNS-AID also uses DNS as a neutral directory

For now, ANS is more of a roadmap than reality. But the Linux Foundation’s decision to pick up the work suggests that agent identity and discovery are poised to become first‑class standards problems, not just features buried in proprietary AI dashboards. As autonomous agents proliferate, the question of “who exactly is this thing I’m talking to?” may prove as important as the models that power them—and ANS is an attempt to answer it in the language the internet already understands: names, keys, and domains.

Noteworthy Linux and open-source stories:

Do NOT Chase SpaceX. Do This Instead.

SpaceX is getting all the attention right now.

NVIDIA, Apple, Tesla, and the other mega-cap names are still dominating the conversation.

But Wall Street’s top-rated analysts are pointing to a different group of stocks.

MarketBeat tracks the highest-rated analyst recommendations every day, and 5 names have just risen to the top.

The Top 5 Stocks to Buy Now report reveals the 5 stocks getting some of Wall Street’s strongest analyst support before the broader market catches on.

If you’re looking for your next move, don’t just follow the names everyone is already talking about.