VulnOps Reimagines Vulnerability Management for AI-Speed Security

| 2 Min Read
VulnOps shifts security toward continuous, AI-led vulnerability response, with policy-based oversight and longer-term defenses that reduce reliance on patching.

VulnOps starts from a problem that traditional vulnerability management was never built to solve: finding flaws is getting faster, while deciding what to fix and safely fixing it remain slow. The emerging model applies AI and automation across discovery, prioritization and remediation, with the aim of making security operations continuous. Its harder question is whether organizations can keep pace without handing disruptive decisions to systems that lack business context.

The concept was outlined by Heather Adkins, Gadi Evron and Bruce Schneier before advanced AI models began surfacing vulnerabilities at what some now call “vulnpocalypse” scale. Since then, the Cloud Security Alliance (CSA) has published an operating model, SANS has issued implementation guidance and plans to introduce training, and security practitioners have begun testing related technologies. The work is moving from an idea toward practice, though the operating boundaries remain unsettled.

VulnOps changes the operating model

The comparison with DevOps is useful because the shift concerns how work gets done, not simply how quickly teams complete an existing process. DevOps brought software development and IT operations closer together through automation and shorter feedback cycles. VulnOps applies a similar logic to vulnerability work: identify issues continuously, make decisions faster and feed the results back into security and engineering workflows.

Rich Mogull, chief analyst at the Cloud Security Alliance, describes that change as moving from an iterative cycle to continuous processes with fast feedback. In his view, vulnerability management needs the same kind of operational redesign that helped teams deliver code more frequently. That makes VulnOps broader than adding an AI feature to a scanner or speeding up a backlog review. It calls for connected workflows that can carry an issue from detection toward an appropriate response.

Ed Skoudis, president of the SANS Technology Institute, sees AI taking on most of the analysis, with people overseeing the operation. That puts agents at the center of the workflow and changes the human role from reviewing every finding to setting rules and intervening when a decision crosses a risk boundary.

Not everyone sees a new discipline emerging. Chris Elgee, a SANS principal instructor and senior security analyst at Counter Hack Challenges, calls VulnOps a possible rebranding of the vulnerability management organizations have always had to do. His skepticism points to a practical test for the idea: teams will need to show that the operating model changes how vulnerabilities are handled, not just how the work is described.

Automation needs a boundary for risk

Remediation is where the promise of automation meets the cost of getting a decision wrong. Applying a patch can require source code changes or updates to third-party packages, and those changes can break established systems. Ali Khan, CISO of Iru, warns that a patch to a long-lived system in maintenance mode could disrupt a business process or interrupt revenue. Security teams therefore have to weigh the vulnerability against the operational risk of a fix.

Mogull says the industry is not ready to automate remediation end to end. In the CSA model, AI can act within predefined policies, while changes with the potential to disrupt operations call for human intervention. The principle is straightforward: give automation authority where the organization has already set acceptable limits, and reserve judgment for changes with broader consequences.

Skoudis describes this as “governance-in-the-loop.” Organizations would define which actions agents can take automatically and where a person must review the decision. For example, an organization might choose to roll out critical Windows workstation patches gradually, updating 25% of devices per hour or two over half a day. The pace and threshold would depend on the organization’s own risk tolerance and operating needs.

Evron, CEO and founder of Knostic, argues that human review may not scale across most vulnerability operations as discovery and exploitation accelerate. He says teams cannot keep a person in every loop and questions whether patching alone can remain the answer. That view is more aggressive than the governance models proposed by Mogull and Skoudis, but it puts pressure on security leaders to decide which decisions truly require human attention—and whether their current processes can make those decisions quickly enough.

Reducing dependence on the patch cycle

For Evron, autonomous patching is a useful step, but it may be only transitional. He argues that defenders need to prepare for conditions in which attackers can move faster than organizations can test and deploy fixes. His proposed direction is to reduce the number of vulnerabilities introduced, contain attacks while systems are running and limit the damage any single flaw can cause.

The first part begins with the tools that produce software. Evron wants security controls built into AI coding agents so that generated code starts from a safer baseline. He also advocates checking code for vulnerabilities as it is generated, rather than waiting for it to reach a conventional continuous integration pipeline. That moves security feedback closer to the point where a flaw enters the codebase.

A second approach is to detect and neutralize exploitation at runtime. Evron calls one version “patch in place during runtime”: identify an attack in system logs and apply a change while the attack is underway. He demonstrated runtime patching at the [un]prompted conference. The proposal shifts some defensive action from a planned update window to the moment a system shows signs of exploitation.

The third approach is defense in depth. If multiple controls reduce the likelihood or impact of exploitation, organizations may not need to treat every critical vulnerability as an immediate patching emergency. This does not remove the need to fix flaws; it aims to give teams more room to respond without leaving a single weakness exposed as the only barrier.

What security leaders can do now

There is a near-term tension between imagining a different security model and getting the basics in order. Evron cautions against assuming that established hygiene alone will be enough as AI changes the pace of vulnerability discovery. Elgee, meanwhile, says many organizations still lack a complete view of their asset inventory. Without knowing what systems they own and where those systems run, teams have a weak foundation for deciding which findings matter or where automated changes should apply.

Khan sees a parallel organizational challenge: security teams cannot keep handing remediation tickets to developers and expect the work to move quickly. Security and engineering need to work more closely, with technical contributions coming from both sides. That requires teams to agree on how to assess risk, test changes and release fixes without treating security as a separate queue.

Skoudis recommends starting within defined governance boundaries rather than trying to transform the whole enterprise at once. Teams can identify low-risk actions agents may perform automatically and set review requirements for decisions with higher business impact. That creates a practical route to automation while giving organizations a way to observe how agents behave before expanding their authority.

Evron also argues that security leaders should reconsider spending on scanning services that produce more findings than teams can address. His point is that discovery is no longer the only bottleneck; remediation is. The more useful investment, in his view, is building trustworthy systems for autonomous patching now while developing defenses that make patching less central over time.

VulnOps will be judged by whether it closes the gap between discovering a vulnerability and reducing its risk. That means connecting security and engineering, granting agents clear authority and keeping a path for human judgment where business disruption is possible. The longer-term challenge is broader still: build systems that can prevent, contain or absorb vulnerabilities when a patch cannot arrive in time.

Source: Christopher Martinez · www.csoonline.com

Comments

Please sign in to comment.