The Linux Foundation Wants Every AI Agent to Carry Something Like a Passport
Digital Transformation

The Linux Foundation Wants Every AI Agent to Carry Something Like a Passport

A new open standard backed by Cisco, Salesforce, and Cloudflare gives AI agents a verifiable identity built on the DNS infrastructure enterprises already run, right as banks start locking outside agents out entirely.

PublishedNovember 14, 2026
Read time6 min read
Share

Agents have been operating without an ID card

Every enterprise identity program assumes you know what is acting on your systems: a user, a service account, a certificate tied to a machine. AI agents have mostly slipped past that assumption. They get spun up inside a workflow, call other services, and disappear, often without a durable, verifiable record of who owns them, what they are authorized to do, or whether the thing calling itself an agent is actually the agent it claims to be.

The Linux Foundation's new Agent Name Service is a direct response to that gap. Announced this week with backing from 11 companies including Cisco, Salesforce, and Cloudflare, ANS defines how agents are named, registered, and authenticated when they interact with each other across organizational boundaries. Linux Foundation CEO Jim Zemlin put the problem plainly: "Autonomous AI agents are transacting and operating across global systems, but they can't succeed without verifiable identity."

Why building it on DNS is the practical choice

ANS is built on top of the Domain Name System, the infrastructure that already resolves every address on the internet and that nearly every enterprise network team already operates, monitors, and understands deeply. Cloudflare CTO Dane Knecht framed the design choice directly: "ANS builds on DNS to help answer that question using infrastructure the Internet already relies on." Reusing infrastructure enterprises already trust, rather than introducing a wholly new layer, is the detail that separates a standard with a real chance of adoption from one destined for a handful of pilot projects and little else.

A proprietary agent-identity scheme would have required standing up new infrastructure, new monitoring tooling, and a fresh set of vendor dependencies just to answer a question enterprises already know how to answer for every other kind of networked entity on their network. Extending DNS instead lets IT and security teams fold agent identity directly into the access control, logging, and audit processes they already run day to day. That single design choice removes most of the integration cost that typically kills identity standards before enterprises ever get around to actually deploying them at scale.

The incidents that made this urgent

Standards bodies rarely move this quickly without a concrete reason, and the reasons behind ANS are recent and specific. The announcement follows a reported attack on Hugging Face involving agents built on OpenAI's models, one of several incidents this year in which autonomous systems acting across organizational boundaries turned into an attack surface instead of the productivity gain they were deployed to deliver. Each incident has fed directly into board-level risk conversations at companies that had previously treated agent deployment as a purely technical decision owned by engineering teams.

Financial services has reacted fastest among regulated industries. JPMorgan Chase CEO Jamie Dimon has said flatly that the bank will not allow outside agents inside its systems, calling them "just too risky." That stance draws a hard perimeter line rather than inviting cautious experimentation with agent governance. A verifiable identity standard gives firms sitting behind that kind of line a credible path toward conditional trust, letting specific, identifiable agents through on a case-by-case basis, provided the standard actually gets adopted widely enough across vendors and platforms to matter to a bank's risk committee.

What adoption actually requires

Eleven corporate backers gives ANS a meaningful start toward real adoption, though a finished rollout across the agent ecosystem remains a considerably longer road. Organizations can adopt or contribute to ANS through the Linux Foundation's project infrastructure starting now. Writing the specification was the easy part of this effort. Getting enough of the agent ecosystem, model providers, orchestration platforms, and the enterprises actually deploying agents in production, to implement it consistently and keep implementing it as their own products evolve is the harder and far more consequential task still ahead.

The agent-to-agent authentication and monitoring pieces of the standard will decide whether ANS becomes real infrastructure or stays a press release every vendor cites without actually shipping support for it. Naming and registration are comparatively easy problems that most identity efforts solve quickly. Proving, at the exact moment one agent calls another system, that it is who it claims to be and authorized to do what it is attempting, is the part every previous identity standard has struggled to deliver at scale, and it is the part enterprise technology leaders should watch most closely over the next few quarters.

The governance opening this creates for CIOs

Most enterprise AI governance policies written over the past two years have focused heavily on model behavior: what an agent is allowed to generate, what data it can access, and which outputs require human review before anything ships to a customer or regulator. Far fewer policies have included a concrete mechanism for answering a simpler, more foundational question first, which is whether a given agent is actually registered, owned, and accountable to a named individual or team inside or outside the organization running it.

ANS gives governance teams a lever they have genuinely lacked until now. A policy requiring every agent touching sensitive data to carry a verifiable ANS identity becomes enforceable through existing network tooling, in a way that a policy merely instructing teams to document their AI use has rarely proven to be in practice. That single change moves agent governance meaningfully closer to the access-control discipline enterprises already apply to every human user and every service account on their network, rather than leaving it as a separate, weaker category of oversight.

The near-term decision for technology leaders

Mandating ANS adoption this quarter would be premature for most CIOs, given how early the standard still is in its rollout. Asking vendors now whether agent identity and naming sit on their roadmap, and whether that roadmap points toward an open standard or a proprietary lock-in mechanism, is a reasonable and useful question to raise in every renewal conversation happening this quarter. The DNS-based approach stands out specifically because it keeps enterprises out of that lock-in trap from the start.

The broader signal here is that agent governance is finally moving from policy documents into enforceable infrastructure, which is where effective governance has always needed to end up to actually change behavior. A rule nobody can technically enforce functions mainly as an aspiration, however well-intentioned the policy document behind it reads. Whether ANS becomes the mechanism that finally makes agent identity enforceable, or one of several competing efforts that end up fragmenting the market instead, will depend less on the elegance of its technical design and far more on how many large platform providers actually ship working support for it in production.

Tagged#news#digital-transformation#enterprise#cio#erp#strategy#governance#linux-foundation#agent-identity#ai-agent-governance#dns-infrastructure#jpmorgan