Identity for machines: the Agent Name Service aims to be the DNS of AI agents

Every time an AI agent acts on behalf of an organization, a question arises that the industry had not yet known how to answer in a standardized way: who really is that agent and what is it authorized to do? The Linux Foundation proposes an answer with the Agent Name Service (ANS), an open standard that extends the Domain Name System to bring verified identity, authentication and discovery to the so-called 'agentic web'. The framework is built around three axes: who an agent represents, what it can do, and whether its code and history remain unaltered.

The smartest thing about the design is what it does not do. Instead of building a blockchain, a new federated registry or a proprietary namespace with its inevitable 'gatekeeper', ANS anchors itself to DNS: an infrastructure with four decades in production, which processes more than 100 million queries per second and in which every registered domain already has a verifiable owner. The agent's identity hangs from the domain the organization already owns. It is an adoption decision, not just a technical one: reusing existing trust is always less friction than manufacturing it from scratch. Support for Decentralized Identifiers and Legal Entity Identifiers adds compatibility with identity systems already deployed under frameworks such as eIDAS.

The backing matters as much as the idea. That GoDaddy —the world's largest registrar— and Cloudflare —operator of a huge share of global DNS traffic— sign on from day one is not a public-relations detail: they are precisely the ones who could turn the standard into real implementation quickly. The presence of Salesforce, Cisco, Infoblox, Hashgraph Online and DistributedApps.ai, whose CEO Ken Huang is the author of the research paper that originated the proposal, suggests a coalition with the weight to push the standard through forums such as the IETF.

It is worth keeping a cool head: for now we are talking about an intent to launch and a standard that will have to earn real interoperability in the field, not just in press releases. The history of the internet is full of identity proposals that never caught on. But the direction is the right one. With 82% of executives planning to adopt agents within one to three years according to the World Economic Forum, the 'shadow AI risk' —agents operating without a neutral mechanism to validate who they are— stops being a hypothesis and becomes a governance problem. Resolving the identity of machines before they multiply is exactly the kind of foresight that tends to be missing in technological waves.