Begin at the End: How to Enable Agentic Remediation
Opinion column urges teams to automate low-risk patching with bounded AI agents and keep humans on high-risk fixes.
A SecurityWeek opinion column argues that continuous threat exposure management still fails at mobilization, where validated findings are ticketed and fixed by hand. It proposes “Shift Zero,” using agentic remediation to apply already-approved patches and configuration changes inside a bounded action space. High-risk exposures would remain human-in-the-loop, while lower-risk changes could run with a human on the loop, rollback plans, and rehearsed failure drills.
- CTEM still stalls at mobilization, where tickets and change windows delay fixes.
- Shift Zero would embed constrained agents that apply known patches and config changes.
- High-risk findings stay human-in-the-loop; lower-risk work can be human-on-the-loop.
- Guardrails include an approved action set, rollback plans, and failure tabletops.
Full article1,006 words · extracted from securityweek.com · click to collapse
There is a corporate mandate to implement AI enablement programs. In cybersecurity, integrating AI chatbots into dashboards is table stakes. The industry, and the organizations it serves, must evolve to adopt agentic AI.
But true autonomy isn’t just about moving faster – it’s about changing the goalpost entirely. The ultimate objective is Shift Zero: eliminating vulnerability backlogs entirely and preventing exposure at the source. To achieve this, the best place to begin is the end.
Continuous threat exposure management (CTEM) is a framework to discover and remediate threats and exposures. The final step of this framework, mobilization, is the operational endpoint. Until now, this has remained a time-consuming manual process.
Automation has historically been focused on discovery, prioritization, and validation. By inverting the traditional CTEM framework to focus on automating remediation, organizations can close the loop on exposure management.
The final frontier for CTEM
Every security team that tries to find and fix its exposures experiences the same lifecycle. You decide what matters. You find out what’s in your environment. You plan your work. You test your plan. You operationalize.
Advertisement. Scroll to continue reading.
This sequence is the CTEM framework: scoping, discovery, prioritization, validation, and mobilization.
The first half of this framework is diagnostic. Scoping defines assets and the attack surface, in relation to their business impact. This is a mostly manual process that relies on human discernment.
Discovery is also diagnostic. Finding out what exists used to be the hard part. Organizations used to rely on static asset inventories, which were maintained in spreadsheets. Scanning is now continuous and automated.
Likewise, prioritization is diagnostic, and it has also been enhanced with automation over the past several years. For a long time, organizations relied on a common vulnerability scoring system (CVSS) to determine the severity of a vulnerability.
Now solutions prioritize vulnerabilities based on whether an exploit for a vulnerability exists in the wild, a compensating control protects an exposure, and the business impact of the assets. Processing all of these variables is an ideal use case for machine learning.
The other half of the framework is action. Validation is the process of confirming that prioritized exposures are exploitable and that controls actually stop attacks. Over the past five years, attack simulation, penetration testing, and adversarial validation have also become increasingly automated.
The final step, mobilization, is where there is the most opportunity to embrace automation. Mobilization operationalizes insights across people and processes.
When you get to this step, the automation stops. Someone opens a ticket.That ticket gets passed across teams, delayed by change windows, negotiated, batched, and occasionally closed without ever being fixed.
We have built an extraordinary solution for producing perfectly prioritized and rigorously validated findings, but the fix still requires manual processes.
CTEM was designed to bring order to exposure management, but manual mobilization has kept security teams trapped in a reactive cycle. Shift Zero flips this paradigm. Instead of managing an ever-growing number of findings, Shift Zero embeds autonomous remediation directly into the operational flow. It takes us from “we found a problem, now let’s fix it” to eliminating the window of risk altogether.
Enabling mobilization with agentic remediation
Agentic remediation is not an act of faith. We are talking about fixing known problems, not judgment calls about unfamiliar risk. An agent applying a known patch to a known asset is executing a decision that has already been made.
We can turn to supervisory control theory, which dictates how human operators control machines they cannot micromanage. This school of thought is responsible for a widely discussed distinction: human in the loop versus human on the loop.
Human in the loop means that agents wait for approval. Agents gather context, determine a fix, stage the change, and stop. A human approves or rejects the fix. The agent does the legwork, while the human supplies the decision.
Human on the loop means that the agent acts and reports. The agent executes its tasks within a defined scope, and a person supervises its output through dashboards and alerts, rather than reviewing individual actions. The human sets policy, and the agent handles the volume.
The natural way to divide work between the two is by the risk score you already produce. Prioritized exposure management rankings are a stronger signal than CVSS severity scores. Use human review on high-risk findings; build confidence in agentic automation with low-risk findings.
Security leaders are reluctant to let an unsupervised agent act on their most impactful exposures, and for good reason. AI systems continue to act in ways that are difficult to anticipate. The fear is that an agent applies a solution that causes further harm.
To allay that concern, agents should operate inside a constrained action space: apply patches from an approved set, modify specific configuration parameters, isolate a segment, and open a ticket for anything else. Destruction isn’t a decision because it isn’t in the vocabulary. Bounding is what makes autonomy safe.
Two more guardrails matter. Every action needs a rollback plan, and approval paths need to be standardized before they’re automated. If ownership of asset classes is ambiguous today, an agent amplifies that ambiguity at machine speed.
Next, rehearse the failure. Prepare tabletop exercises to practice an agent patching the wrong host at the wrong hour. Find out who notices, who has the authority to stop it, and how long it takes. Those lessons are better learned in a conference room than a production environment.
In the future, the promise of agentic remediation is even more ambitious: self-healing networks, or networks that monitor their own telemetry, model normal behavior, and isolate anomalies autonomously.
CTEM is supposed to be a closed-loop lifecycle. Discovery, prioritization, and validation have been running at machine speed for years, but mobilization has been a manual process, which means the loop gets left open.
Agentic remediation is where the mandate to adopt AI meets a problem worth solving. The exposures are already found, ranked, and validated. What remains is the fixing. Begin at the end, close the loop, and Shift Zero becomes operational reality.