COMMENTARY: In a classic joke, two hikers encounter a starving bear. One hiker drops to the ground to pull sneakers from his backpack. “You can’t outrun a bear!” his incredulous companion shrieks. “I don’t have to outrun the bear,” the first hiker replies as he shoves his feet into the shoes. “I just have to outrun you.”
For years, we’ve been engineering better and faster sneakers for our users. Multifactor authentication, adaptive access, and passkeys have made it exponentially harder for attackers to compromise
human identities. And they’re effective at blocking thousands of attacks per second.
[
SC Media Perspectives columns are written by a trusted community of SC Media cybersecurity subject matter experts. Read more Perspectives here.]
But with this significant progress comes an uncomfortable truth: the bear is still hungry. And it's now chasing the other hiker’s workloads.
The shift we never saw coming
Twenty years ago, we built server software that assumed anyone inside the building and connected to the network belonged there. Putting a lock on the door kept the bears away — until we started connecting everything to the internet.
Today, we’re deploying workloads — software that runs on servers and serves multiple users — faster than we're hiring employees. People are already writing
AI agents that can write their own agents. With this exponential acceleration comes the harsh reality that most workloads are weakly authenticated, dangerously over-privileged, and poorly governed.
Consider a seemingly innocent
AI-powered agent for legal discovery that has "read-only" permissions to all employee mailboxes. If it gets compromised, an attacker suddenly has direct access to the organization's internal communications — meeting invites, confidential discussions, sensitive attachments. And because the assistant was granted access by design, its activity probably won't trigger any alarms.
This isn't a hypothetical scenario. It’s very real.
Why the bear chases the workloads
The economics are simple. When it was easy to hack the passwords or phish the credentials of 100 employees, why bother attacking a much harder-to-reach admin system? But now that we've raised the security bar on user authentication, the math has changed. Individual IT or sys admins and the workloads those admins deploy have become far more attractive. Compromising one workload gives the attacker access to thousands of users' data — without trying to outmaneuver modern identity security to compromise thousands of user accounts.
There are two primary attack vectors. First, workloads themselves may have security vulnerabilities (a compromised password) that an attacker can exploit via the internet. Second, attackers are getting bold — from malware installed locally on admin machines to good old-fashioned blackmail or even threats of bodily harm. Once they compromise an IT admin, they gain access to workloads running with broad privileges across your infrastructure.
The trust model between workloads is often implicit and
overly permissive. Unlike users who change roles and undergo regular access reviews, workloads run continuously with the same elevated privileges granted to them on day one. They become invisible highways for lateral movement that attackers can use to exfiltrate data at scale.
Zero-trust isn't just for humans
We've become diligent about applying
zero-trust principles to users: verify explicitly, enforce least privilege, and assume breach. It's time to apply that same rigor to our workloads. Here’s what teams need:
Strong workload authentication: The easiest wins come from eliminating passwords and secrets between workloads. If a workload authenticates to another service using a password stored in code or configuration, the company is one leak away from disaster. All major cloud providers offer managed identity services — use them. On-behalf-of token models: Stop giving workloads blanket access to everything. When an application frontend talks to the backend, it should not only prove its own identity, but also specify which user it's acting for. The backend should then grant access only to that user's data. This does require engineering work, but OAuth standards already support these patterns, and the investment pays off by drastically reducing the team’s potential blast radius. Governance and continuous review: Just as the team regularly reviews user access, it also needs to govern workload permissions. Which services can talk to each other? Are those permissions still necessary? Unlike the previous two measures, this has become an ongoing discipline, not a one-time fix. The good news: tools are available to help. The bad news: many organizations don’t understand their value, so too few have adopted them. Don’t feed the bear
We're witnessing an emerging flood of smart, multi-functional, and
autonomous AI agents that will soon handle everything from engineering specifications to marketing campaigns with minimal human interaction. This means the next major breach won't start with a stolen password. It'll start with misplaced trust — trust we've inadvertently built into the invisible connections between our systems.
No one can afford to wait until that breach happens to take action.
Teams need to prepare for this new reality: Keep tabs on the company’s workloads at all times. Inventory all workload identities. Map their permissions. Apply zero-trust principles consistently.
In this race, get those sneakers ready because there’s no prize for second place: only consequences.
Joy Chik, president, identity and network access, Microsoft SC Media Perspectives columns are written by a trusted community of SC Media cybersecurity subject matter experts. Each contribution has a goal of bringing a unique voice to important cybersecurity topics. Content strives to be of the highest quality, objective and non-commercial.
