AI is changing cybersecurity on both sides of the equation.
Defenders are using AI to find vulnerabilities faster. Attackers are gaining new ways to analyze disclosures, develop exploits, and compress the time between discovery and attack. Meanwhile, security teams are already managing more vulnerabilities than they can realistically investigate or remediate.
New research from Google Threat Intelligence Group makes the scale of that shift difficult to ignore.
In its September 30 report, “Vulnerability Discovery and Exploitation Trends in the AI Era,” Google found that vulnerability disclosures more than doubled in 2026, climbing from 5,045 in January to 10,740 in August. At the same time, the average number of vulnerabilities observed being exploited in the wild increased from 10.5 per month in 2025 to 18 per month during the first eight months of 2026.
Those numbers are significant.
But one of the most important findings in the report is not how many vulnerabilities exist.
It is how few actually become part of an attack.
Google found that only 0.23% of disclosed vulnerabilities in 2026 were observed being actively exploited, roughly one in every 431 vulnerabilities.
That distinction should fundamentally change how organizations think about vulnerability management.
More Vulnerabilities Do Not Automatically Mean More Risk
For years, vulnerability management has largely operated as a numbers problem.
Find vulnerabilities. Assign severity. Patch as many as possible. Measure progress by how quickly the backlog goes down.
That model becomes increasingly difficult to sustain when vulnerability discovery itself is accelerating.
If AI enables researchers, vendors, defenders, and eventually attackers to find weaknesses at a dramatically faster rate, trying to patch everything faster becomes a losing race.
Google’s findings illustrate exactly why.
High-risk vulnerability disclosures increased from 131 in January 2026 to 350 in August, according to GTIG. But even as vulnerability volumes increased dramatically, exploitation remained concentrated in a comparatively small subset of weaknesses.
The question security teams need to answer is therefore not:
How many vulnerabilities do we have?
It is:
Which exposures can an attacker actually use to reach something that matters?
That is an exposure management problem.
AI Is Compressing the Defender’s Decision Window
AI introduces another complication.
Google’s research suggests AI-assisted vulnerability discovery is not simply producing more findings. The vulnerabilities identified with AI assistance appear to have a different risk profile.
Among the vulnerabilities Google could identify as likely AI-discovered, 58% were rated Medium Threat Risk compared with 28% of vulnerabilities not identified as AI-discovered. AI-assisted discovery also produced proportionally fewer low-risk findings. Google cautions that the data is still an early indicator, in part because there is no standardized method for identifying AI-discovered vulnerabilities.
At the same time, attackers may be able to use AI to accelerate what happens after a vulnerability becomes public.
Google points specifically to the potential for AI and large language models to help threat actors analyze patches, compare software versions, examine proof-of-concept code, and weaponize known vulnerabilities faster. The report attributes much of 2026’s increase in exploitation to the rapid weaponization of these n-day vulnerabilities rather than to a dramatic surge in zero-day attacks.
As we explored in The Asymmetry Problem, AI is changing the economics of attack by helping adversaries move faster while defenders still have to make careful decisions about what to investigate, prioritize, and remediate.
That makes the defender’s decision window even smaller. Knowing that a vulnerability exists is no longer enough.
Organizations need to quickly understand:
- Is the affected asset exposed?
- Can an attacker reach it?
- What security controls stand between the attacker and the asset?
- Could exploitation create a path to a critical system?
- What would the business impact be if that path were successfully traversed?
- What change would reduce the most risk without unnecessarily disrupting operations?
Answering those questions requires context that a vulnerability record alone cannot provide.
The Next Security Bottleneck Is Context
This is where vulnerability management and exposure management begin to diverge.
A vulnerability scanner can tell you that an asset contains a weakness.
It cannot, by itself, tell you everything an attacker could do with that weakness inside a complex hybrid environment.
That distinction becomes even more important as enterprise environments span on-premises infrastructure, public and private clouds, remote users, operational technology, IoT devices, identity systems, security controls, and third-party infrastructure.
A vulnerability that looks critical on paper may be isolated behind effective controls.
Another vulnerability with a lower traditional severity score may sit on an exposed asset with a viable path to a critical business system.
Treating those two findings the same wastes valuable remediation capacity.
Security teams need to understand vulnerabilities as part of the environment around them.
Attackers Think in Paths, Not Lists
Google’s findings about edge infrastructure reinforce this point.
Edge and security appliances represented 14% of vulnerabilities exploited from January through August 2026, and more than 65% of exploited edge vulnerabilities carried High or Critical Threat Risk ratings. Google notes that adversaries continue to target exposed management interfaces as initial access points.
But gaining access is only the beginning of an attack.
Once an attacker gets in, the next questions are the same ones defenders should be asking beforehand:
Where can I go?
What can I reach?
Which controls can I bypass?
Which assets give me access to something more valuable?
This is why exposure management cannot stop at vulnerability severity.
Organizations need to understand potential attack paths across the environment and determine which combinations of vulnerabilities, configurations, connectivity, identities, and security controls could allow an attacker to move from an initial foothold toward critical assets.
The goal is not simply to find the vulnerability before the attacker.
It is to understand the environment better than the attacker does.
AI Infrastructure Creates New Exposure, Too
Google’s research also highlights another emerging challenge: AI infrastructure itself is becoming part of the attack surface.
The report found growing vulnerability disclosures across agent orchestration frameworks, AI gateways, inference infrastructure, and related middleware. Agent orchestration frameworks accounted for half of the AI-related vulnerabilities examined by GTIG, while vulnerabilities affecting inference infrastructure included issues involving unauthenticated APIs and server-side request forgery.
This matters because AI infrastructure does not exist in isolation.
It connects to APIs, databases, cloud resources, development environments, credentials, internal applications, and increasingly sensitive enterprise data.
A compromised AI gateway or orchestration platform may therefore represent more than a vulnerable application. It can become another entry point into the broader enterprise environment.
Security teams need visibility not only into the vulnerability itself, but into what that system connects to and what an attacker could reach next.
Exposure Management Turns Vulnerability Data Into Decisions
This is where RedSeal’s approach to preemptive exposure management becomes increasingly important.
RedSeal builds a continuously updated model of the hybrid environment across on-premises, cloud, remote, OT, and IoT infrastructure. That model provides the foundation for understanding assets, connectivity, controls, exposures, and potential attack paths across environments that are often managed through separate tools.
From there, organizations can move beyond vulnerability counts and answer the questions that actually drive risk reduction.
See the environment as an attacker could
RedSeal’s Hybrid Environment Modeling creates a living view of assets, connections, and controls across the hybrid enterprise. That visibility helps teams understand where vulnerable assets sit within the broader environment rather than evaluating them in isolation.
Determine whether an exposure creates a viable attack path
RedSeal’s Attack Path Analysis models the ways threats could move through the environment and evaluates the security controls along those routes. Instead of assuming every vulnerability represents the same level of danger, teams can identify the paths that create meaningful exposure to critical assets.
Prioritize based on risk, not vulnerability volume
With Risk Radius™, RedSeal combines exploitability with business impact to help security teams identify which exposures should be addressed first. This shifts remediation from simply working through a vulnerability queue toward reducing the exposures that represent the greatest risk to the organization.
Give teams evidence they can act on
Exposure management ultimately has to result in action. That makes CTEM feel like a natural extension of the argument instead of something bolted on for lead gen. RedSeal provides SecOps and NetOps teams with evidence about affected assets, connectivity, attack paths, and controls so remediation can focus on specific changes rather than sending another undifferentiated list of vulnerabilities to infrastructure teams.
The Future of Vulnerability Management Is Not More Patching. It Is Better Prioritization.
Google concludes that organizations need to move away from unprioritized mass patching and toward risk-based triage as vulnerability discovery and exploitation accelerate.
We agree.
AI will continue making vulnerability discovery faster.
It will make vulnerability analysis faster.
It will increasingly help defenders remediate problems faster.
And attackers will benefit from many of those same advances.
That means speed matters. But context determines where that speed should be applied.
The security teams that succeed in this environment will not necessarily be the organizations that patch the most vulnerabilities.
They will be the organizations that can continuously understand their hybrid environment, identify the exposures that create viable paths to critical assets, prioritize them according to actual business risk, and mobilize remediation before those paths are exploited.
That is the shift from vulnerability management to exposure management.
And in the AI era, that shift is becoming increasingly urgent.
Ready to put exposure management into practice?
Learn how Continuous Threat Exposure Management helps security teams move from reactive vulnerability management to a repeatable process for identifying, prioritizing, validating, and reducing the exposures that matter most.
