We have never been short of vulnerabilities. We have always struggled with remedial work.
For decades, the cybersecurity industry has approached vulnerability management from the wrong end. The emphasis has consistently been on finding vulnerabilities, cataloguing them, scoring them and, more recently, prioritising them through increasingly sophisticated risk models.
An entire industry has developed around vulnerability scanning, exposure management, threat intelligence and remediation prioritization. Artificial intelligence is now promising to take those capabilities to another level.
But there is something fundamentally wrong with that approach.
The real challenge in vulnerability management has never been identifying what needs fixing. It has always been getting the remedial work done.
Most large organizations already know they have significant numbers of vulnerabilities across their technology estates. They have known for years. They have dashboards, reports, inventories, risk ratings and remediation backlogs.
What they frequently lack is the organizational machinery to translate that knowledge into effective, sustained remedial action.
And that is not primarily a technology problem.
The uncomfortable reality behind remediation backlogs
Consider what happens when a large enterprise identifies a critical vulnerability.
The security team assesses its severity and potential impact. It raises a ticket, assigns a priority, identifies an owner and requests remediation.
On paper, the process looks straightforward.
In practice, that is often where the real problems begin.
The infrastructure team may be managing ageing systems with complex dependencies. The development team may have committed to a demanding release schedule. Production operations may be unwilling to accept additional downtime. Business stakeholders may resist disruption to revenue-generating services.
Some systems may have very limited maintenance windows. Others may require extensive regression testing before deploying patches. Legacy applications may depend on components that cannot easily be upgraded.
And ownership itself may be disputed.
Security teams frequently have the responsibility to identify and report vulnerabilities but lack the authority to enforce remediation. Technology teams control the systems but may have competing priorities and insufficient resources.
The result is familiar: escalating tickets, repeated exceptions, missed deadlines, unresolved disputes and growing remediation backlogs.
None of these problems can be solved simply by producing a better vulnerability score.
Prioritization is necessary, but it is not the answer
The industry has increasingly recognized that organizations cannot remediate everything.
That has driven the development of risk-based vulnerability management, exploitability intelligence, asset criticality models and sophisticated exposure prioritization platforms.
These capabilities have genuine value: There is an obvious difference between a vulnerability that is actively exploited on an internet-facing critical system and one that exists on an isolated, low-value asset.
But prioritization is only the beginning of a management process. It is not its outcome: A perfectly prioritized remediation backlog remains a backlog.
In many organizations, the difficulty is not deciding which vulnerabilities matter most. It is securing agreement on who will fix them, when the work will take place and who has the authority to resolve conflicts or accept the risk.
These are matters of governance, accountability and business priorities.
The cybersecurity industry remains disproportionately focused on the what, when the real challenge lies in the how and the who.
AI is about to make the problem considerably worse
Artificial intelligence is fundamentally changing the economics of vulnerability discovery.
AI-assisted tools can analyse code, identify weaknesses and accelerate vulnerability research on a scale and at a speed that would previously have required substantial human effort.
According to Google Threat Intelligence Group, monthly vulnerability disclosures more than doubled between January and August 2026, while observed exploitation also increased significantly. AI is contributing to that acceleration, although changes in disclosure practices also influence the figures.
The consequences for large organizations could be significant.
We are moving towards a world where the identification of vulnerabilities becomes increasingly automated, increasingly inexpensive and potentially continuous.
But what happens when the volume of identified vulnerabilities increases dramatically while the organization’s capacity to remediate them remains essentially unchanged?
The answer is predictable: Larger backlogs. More pressure on technology teams. More exceptions. More disputes over priorities. And potentially greater exposure, despite unprecedented visibility into technical weaknesses.
The Cloud Security Alliance has already highlighted this growing disconnect between accelerated vulnerability discovery and the capacity of organizations to absorb and remediate findings.
AI is industrialising vulnerability discovery while remediation remains trapped in organizational structures designed for another era.
That is the real emerging challenge.
AI cannot negotiate a production maintenance window
Of course, AI can also contribute to remediation.
It can establish proofs of concept, generate and test fixes, propose configuration changes, analyse dependencies and automate certain deployment activities.
These are important developments, and organizations should embrace them where appropriate.
But their practical impact must not be overstated.
AI cannot independently decide that a critical production system should be taken offline during a peak business period. It cannot assume accountability for disrupting customer services or resolve competing priorities between a business unit, an infrastructure team and the security function.
It cannot simply override legacy dependencies, contractual obligations or regulatory constraints.
Even where AI can technically generate a suitable fix, somebody still needs to validate it, approve its deployment, ensure that the fix has not broken something else and verify that remediation has been successful.
Some of those steps can be automated. Others require judgement, authority and organizational coordination.
The more powerful AI becomes, the more visible those remaining management bottlenecks will be.
And this is where the traditional cybersecurity approach is increasingly inadequate.
Vulnerability management needs an operating model, not another dashboard
What large organizations need is not simply better vulnerability management technology.
They need a remediation operating model that establishes clear accountability across security, development, infrastructure, operations and business management.
That model should address several fundamental questions.
Who owns the remediation of vulnerabilities affecting shared platforms or complex applications? Who allocates the necessary resources? Who can authorize emergency changes? Who resolves disagreements when remediation conflicts with business priorities?
How are maintenance windows negotiated and secured? How are persistent exceptions challenged? Who accepts the residual business risk when remediation cannot be completed within the expected timeframe?
And, crucially, who is responsible for ensuring that agreed actions happen?
These responsibilities cannot be left to informal relationships between technical teams or to the persistence of individual security managers.
They need to be embedded in management processes, supported by executive authority and reinforced through meaningful performance measures.
Security should provide risk intelligence, challenge and assurance. Technology teams should own the delivery of remediation within their environments. Business management should participate in decisions where operational disruption or investment trade-offs are unavoidable.
The CISO’s role is to help establish and oversee that machinery, not to become the organization’s chief ticket chaser.
Measuring remediation effectiveness, not just reporting activity
There is also a fundamental problem with how vulnerability management performance is measured.
Too many organizations remain focused on the number of vulnerabilities identified, severity distributions, compliance with remediation deadlines and the volume of outstanding findings.
Those indicators have their place, but they can encourage an administrative approach to vulnerability management: Closing tickets becomes a proxy for reducing risk.
What matters is whether material exposures are being eliminated within acceptable timeframes and whether the organization can demonstrate that remediation has worked.
Management should pay particular attention to recurring remediation failures, ageing high-risk exposures, operational bottlenecks, the effectiveness of exception handling and the time required to implement verified fixes.
The distinction between a vulnerability being reported as resolved and genuinely being remediated is essential.
AI can help generate more accurate and timely evidence. But it cannot substitute for accountability over the results.
Auditability and regulatory obligations add another critical dimension to this challenge.
Regulators increasingly expect organizations not only to identify and assess vulnerabilities, but also to demonstrate that appropriate remedial actions have been taken within reasonable timeframes.
This requires clear ownership, documented decisions, traceable exceptions periodically reviewed and reliable evidence that remediation has been completed.
As AI accelerates vulnerability discovery and potentially automates parts of the remediation process, maintaining an auditable chain of accountability becomes even more important.
Ultimately, compliance cannot rest on the ability to identify problems: It needs to rest on the ability to demonstrate that they have been effectively addressed.
The real opportunity for AI is to transform execution
This is where I believe the cybersecurity industry should concentrate its attention.
Rather than simply applying AI to discover more vulnerabilities and refine their prioritization or their treatment, vendors and practitioners should focus on how AI can accelerate the entire remediation lifecycle.
That means understanding dependencies, identifying suitable deployment opportunities, preparing changes, automating testing, coordinating workflows, validating results and providing management with reliable evidence of progress across the whole cycle.
In highly automated environments, AI agents may eventually execute substantial portions of that process under clearly defined controls.
But the objective should not be automation for its own sake: It should be to reduce the friction between identifying a problem and resolving it.
And that requires technology to be designed around a functioning governance model, rather than expecting governance to emerge from the technology.
The technology will change. The management problem will not.
I have argued for years that cybersecurity is fundamentally a leadership and management challenge supported by technology.
Vulnerability management provides perhaps one of the clearest demonstrations of that principle.
The same structural problems that prevented organizations from patching critical systems fifteen years ago remain visible today: fragmented responsibilities, competing priorities, operational constraints, weak governance and insufficient management follow-through.
AI will not make those problems disappear.
On the contrary, by accelerating vulnerability discovery and shortening the time available to respond, it is exposing them more brutally than ever.
The organizations that succeed will not necessarily be those with the most sophisticated vulnerability discovery capabilities.
They will be those that have built the governance, accountability and operational discipline required to translate technical intelligence into effective remedial action.
The future of vulnerability management will not be determined by how quickly we can find what is wrong. It will be determined by how effectively organizations can get things fixed.
And that has always been a management problem.