Authenticating workloads is becoming more and more complex, particularly given things like AI agents and the wide range of identity permissions they need. Organizations need to be thinking ahead on securing workloads in complicated modern environments, but it’s not an easy task.
Researchers at Zscaler hope to explore this evolution in an upcoming RSAC 2026 Conference session entitled, “What Are You, Really? Authenticating Workloads in a Zero Trust World.”
In computing terms, workloads cover the tasks applications and services conduct in order to do their job, and the IT resources those tasks consume. Workloads can refer to a wide range of things, from processing front-end user requests on a Web server (like managing a shopping cart) to cloud-native microservices, complex data analysis, AI training, and more.
The Challenges of Tackling Workloads in 2026
Many workloads conduct their tasks quietly in the background and are considered non-human identities (NHI) because they require permission and authentication, much like human IT personnel would.
When you consider AI agents, which attempt to emulate the job a human might do, down to autonomous reasoning and decision-making (to whatever extent an agent can), the workloads get more complicated and require more stringent security controls. Also, especially in large companies, they could be using Azure, Google Cloud, and AWS to meet different needs, alongside on-premises services. Organizations need to authenticate workloads in a way that scales across the different aspects of a given environment.
During their upcoming technical session, Zscaler chief information security officer (CISO) Sam Curry and chief scientist Yaroslav Rosomakho will cover multiple specific methods for authentication, such as the mutual TLS (mTLS) security protocol, workload identity tokens, and remote attestation, as well as offer specific insights into which methods scale better than others.
Rosomakho tells Dark Reading that, historically, workload authentication and identity were not top of mind for organizations, and that while earlier on “it was a simpler world,” things have quickly grown complex. That complexity, unfortunately, doesn’t match the way many organizations currently secure their non-human identities.
“What we observe is that, right now, there are widespread insecure practices when it comes to workload identity,” the chief scientist says. “In many organizations, they simply rely on static IP addresses for identity mapping, and obviously that scales poorly. It’s spoofable, and any change to infrastructure collapses your workload identity definitions. We also see plenty of organizations that rely on all sorts of static credentials, such as HTTP basic authentication.”
Moreover, Rosomakho says the most common way organizations identify and authenticate AI agents specifically is through static headers and keys that are never rotated.
“It’s a significant problem,” he says, adding that tying important processes to static keys can be a recipe for major technical and financial damage against an unprepared defender.
How to Authenticate Workloads in Your Environment
Curry tells Dark Reading that, from a defender standpoint, there are many options to solve these problems and remediate the weaknesses. At a basic level, he says organizations should be looking for secrets, taking inventory of AI agents (as well as other NHI processes and services), adopting standards, and working toward zero-trust. They should also be talking to their platform providers about also adopting workload authentication standards.
“It’s about testing federation and defining [a data security] policy,” he explains.
All of that said, the appropriate defense posture does depend on what the organization’s specific needs are. For example, Kubernetes Service Accounts make it so that workloads spun up in Kubernetes get dynamic short-term identities and can authenticate themselves to the outside world safely.
An organization may alternatively or additionally want to consider adopting one of the many open source standards that exist for this exact purpose, such as Secure Production Identity Framework for Everyone (SPIFFE), which, according to its website, is used “for securely identifying software systems in dynamic and heterogeneous environments.” At the heart of SPIFFE, as well as many of the better solutions, is creating a well-defined environment built on short-lived identities.
There’s also the Internet Engineering Task Force’s Workload Identity in Multi-System Environments working group, or WIMSE. WIMSE focuses primarily on defining standardized solutions for tackling the many problems that come up in addressing workloads today. They have meetings, a charter, a mailing list, and relevant documents.
Whether an organization wants to adopt either of these standards or another like Security Assertion Markup Language (SAML), Curry and Rosomakho argue in favor of taking steps now, as workloads show no sign of getting less complex.
“It’s arguable that the most interesting and most common and most valuable communications that will be happening in our economy are going to involve no humans,” Curry says. “And so, it behooves us to be able to apply confidentiality, integrity, and availability in those circumstances. We can’t do that without a more advanced schema for authentication and then authorization. It might be one of the most important subjects for people in the cyber world or the IT world to say, OK, what’s our strategy here?”
Source: www.darkreading.com…