|
There has been a lot of talk lately about agent identity – the idea that, as AI agents increasingly come into use, it’s becoming critical to have some way of identifying which people and/or organizations control them. In the identity technology industry, a number of organizations have rushed to address the issue, pitching their own solutions for tying a given agent to a specific person. That remains a key issue to address – right now there are a lot of abstract solutions competing for attention – but in the meantime, there’s a whole other side to this question that may now have a key solution: open registries of known AI agents.
A couple of weeks ago, the Linux Foundation announced its intention to champion the Agent Name Service (ANS) as an open standard that ties agent identities to existing Domain Name Service infrastructure. In other words, its idea is to tie an agent directly to a registered website domain.
For example, let’s say General Motors created a procurement agent that would go out and order materials semi-autonomously. GM could register the agent under ANS, perhaps using the name “procurement-agent.gm.com”. Then, when the agent contacts some other company – a metal supplier, for example – that company can look it up via the ANS and confirm that it really is an officially registered agent under the General Motors web domain.
This solution has taken shape pretty quickly. It first emerged in a research paper – Agent Name Service (ANS): A Universal Directory for Secure AI Agent Discovery and Interoperability – published in May of last year. That October, GoDaddy – the popular domain registrar – picked it up and ran with it, announcing a new service that “draws on concepts documented in an Internet Engineering Task Force (IETF) draft for an Agent Name Service.” GoDaddy made some important architectural changes, and saw considerable adoption as the project was rolled out, with thousands of third party agents registered by April of this year.
Now, the Linux Foundation is looking to make this an open standard rather than something that is just hosted by GoDaddy. The plan now is to open the technical work to broader contributors through the ANS GitHub organization and standards process, and to enable various companies to run their own ANS-compatible systems, which will all follow the same rules and be interoperable to some degree.
From here, a key question will be adoption. Cisco, Salesforce, Cloudflare, Infoblox, and, of course, GoDaddy have all publicly come out in support of this standard, so that’s promising in terms of network effects.
The upshot is that agent provenance seems like it could be solved here – again, depending on adoption. ANS doesn’t just check for some kind of agent-related subdomain: It actually binds a given agent to a registered domain through cryptographic credentials. So a third party can be reasonably sure whose agent it is interacting with.
There remain many other questions to answer. The ANS won’t necessarily say exactly what a given agent is authorized to do, or who authorized it. It can’t attest that the agent really is allowed to be doing any particular transaction. And the agent’s registration in itself does not tie the agent to any particular people, so it can’t really say who is responsible for its actions. But it’s a very important start.
If the ANS idea does indeed take hold, it will establish infrastructure that a lot of organizations will depend on, and that identity solutions providers will want to leverage – especially if they think they have the answers to some of the important remaining questions.
|
