Deep dive

The Patch Window Is Stable on Average. The Dangerous Cases Aren’t.

October 26, 2026

The Patch Window Is Stable on Average. The Dangerous Cases Aren’t.

For builders, founders, and security leaders, the practical question is not whether vulnerabilities are disclosed faster. It is whether your team can still patch the internet-facing, high-severity ones before attackers turn them into working exploits.

That distinction matters. The average patch window can look stable while the dangerous tail compresses to hours or days. If you plan around the average, you can still lose on the exposures that matter most.

Two different problems are getting blurred together

A lot of bad analysis comes from mixing up two timelines:

  • disclosure-to-exploit: how quickly attackers weaponize a flaw after it becomes public or known
  • disclosure-to-patch: how long it takes vendors and operators to ship and deploy a fix

Those are not the same thing. A CVE can be disclosed, exploited, and only later patched. Or it can be patched quickly but exploited before most customers have applied the fix. When those timelines get collapsed into one chart, the result can look more dramatic — or more comforting — than reality.

That is why the patchwindow analysis is useful. It argues that some “collapse” charts are artifacts of measurement, not proof that all exploitation is instantaneous. In its framing, the like-for-like mean patch window has stayed fairly steady, and elapsed-time graphs can make old CVEs look slower than they really were. Its warning is blunt: “Any per-year statistic that doesn’t control for elapsed time will mechanically decline year over year, regardless of what attackers are actually doing.” 1

Brandefense makes the same methodological point from another angle: “Publication is not the start of exploitation. A CVE can be exploited before its record exists, or long after. The delay measures a catalog event against a record date, not attacker activity.” 2

That caveat should not be read as reassurance. It is a reminder that the public record is only a proxy for the real-world threat timeline.

The dangerous tail is moving much faster

The strongest sources in this set do not say every vulnerability is now a race against the clock. They say that high-severity, internet-facing, and otherwise attractive exposures are being exploited much faster than many organizations can remediate them.

Rapid7 reports that in Q3 2026, the average time between disclosure and exploitation for high-severity vulnerabilities was three days, with 71% exploited within seven days and 90% within 30 days. 3 Microsoft’s 2026 Digital Defense Report pushes the same point harder: the median time from discovery in the wild to weaponization has fallen below 24 hours, while enterprise remediation for critical external vulnerabilities still takes 30 to 60 days. 4

That is the gap that matters for operators. If your patch process still assumes triage, testing, change approval, and rollout will fit inside a normal business cycle, you are already outside the attacker’s timeline for the cases they care about.

Microsoft is explicit about the consequence: “That mismatch gives threat actors a widening window to act before exposed systems are fixed.” 4

“The mean patch window has remained remarkably steady at an average of 52.5 days. The 2026 prediction lands at 47.5 days, marginally below the current average.”

— patchwindow 1

That line is easy to misread as comfort. It is not. It says the average patch window is not collapsing across all CVEs. It does not say the internet-facing, high-severity cases that attackers choose are safe.

Where the gap is widest

The gap is biggest where patching is hardest and attacker payoff is highest: edge devices, exposed services, supply-chain infrastructure, and widely deployed software that is painful to update.

Google Cloud’s M-Trends report says the mean time to exploit vulnerabilities dropped to an estimated -7 days, meaning exploitation is routinely occurring before a patch is even released. 5 That is a strong claim, but it should be read in context: it describes the most serious exposures and the way attackers prioritize them, not some universal state for every bug in every product category.

Rapid7’s report supports that narrower reading. It found that 79% of the newly exploited high- and critical-severity vulnerabilities it tracked in Q3 2026 required no authentication or user interaction. 3 Those are the kinds of flaws that can be operationalized quickly. They do not need a slow, elaborate kill chain to matter.

The lesson for product teams is straightforward: if your remediation model depends on user action, discretionary downtime, or a routine maintenance cadence, it is probably too slow for the exposures attackers are actively selecting.

AI is compressing the attacker workflow, but not magically

The more grounded claim in the evidence is not “AI invents more zero-days.” It is that AI lowers the cost of turning public information into working exploit code and accelerates the path from disclosure to weaponization.

McKinsey says the average time between disclosure and active exploitation for critical exposures has fallen to a matter of hours, down from roughly three weeks in 2025. 6 Cogent Research describes the mechanism more concretely: LLM-based tools can ingest a patch diff, identify the security-relevant code change, and generate proof-of-concept exploit code in hours rather than weeks. 7

That is the practical shift for builders and investors. The bottleneck is no longer just “Can someone find the flaw?” It is “How quickly can a public disclosure become a usable exploit chain against a real target?”

VulnCheck adds an important correction to the hype cycle: the data still suggests AI-assisted discovery is overhyped relative to the evidence, and the bigger change is faster exploitation of known weaknesses rather than some flood of magical new vulnerabilities. 8 That is the right place to focus attention. The likely risk is throughput, not science fiction.

A separate 1 Minute Signal analysis of an a16z discussion makes the same directional point in plainer language:

“AI is not just discovering vulnerabilities; it is effectively 'autonomizing' the exploitation of existing, known weaknesses in the software supply chain.”

— 1 Minute Signal coverage of a16z 9

The value of that framing is not that it proves a new universal law. It is that it captures how attacker workflows are being shortened around known bugs, credential theft, and supply-chain weaknesses.

Disclosure infrastructure is also the bottleneck

The widening gap is not only an attacker problem. It is also a disclosure-system problem.

Cloud Security Alliance argues that AI-generated vulnerability findings are stressing coordinated disclosure processes and public enrichment systems. It says the gap is widening “in both directions: faster finding, slower absorption,” and warns that the public record can underrepresent the attack surface when assignment and enrichment cannot keep pace. 10

That matters because many teams still rely on public records as if they were a complete map of risk. They are not. If discovery outpaces CVE assignment, vendor triage, and enrichment, then the public record can lag behind what defenders actually need to know.

Brandefense’s distribution data shows why simplistic planning assumptions fail. Its KEV-to-CVE delays have a 25th percentile of 1 day, a median of 20 days, and a 75th percentile of 259 days. 2 Some exposures move almost immediately; others remain unresolved for months. A single median obscures that spread.

That is also why the disclosure-to-patch frame needs to stay separate from disclosure-to-exploit. A bug may be public, but not yet patched. Or patched, but still widely exploitable because adoption lags. Security teams have to manage both timelines at once.

A concrete example of why timing now drives decisions

The operational implication is not abstract. It shows up in how teams should decide what gets emergency treatment.

A recent 1 Minute Signal coverage of David Ondrej’s reporting on the OpenCode incident describes an attack linked to AI-assisted resource theft and missing funds on the victim side. The broader point is that the harm can show up quickly, and it can hit smaller platforms and individual users, not just large enterprises.

“The OpenCode incident serves as an early indicator that AI-driven resource theft is moving downstream to individual users and smaller platforms.”

— 1 Minute Signal coverage of David Ondrej 11

That is a useful reminder for founders. If you run a smaller product or a fast-moving platform, you should not assume you are below the threshold of attacker attention. Smaller targets can be attractive precisely because they move more slowly, have thinner controls, or expose a simpler path to monetization.

What builders should do with this

The point is not “patch faster” as a slogan. It is to assume that for the exposures attackers care about — especially internet-facing and high-severity ones — disclosure may become attacker material almost immediately.

That changes the operating model:

  • Treat critical internet-facing vulnerabilities as urgent by default.
  • Build pre-approved response paths for emergency patching, segmentation, and temporary mitigations.
  • Pair patching with detection and hardening, because patch release alone may not beat the exploit curve.
  • Prioritize credential and edge-device exposure, since those remain easy entry points even when the vulnerability count is large.
  • Separate disclosure-to-patch from disclosure-to-exploit in your internal reporting, so remediation metrics do not hide active attacker pressure.

The broader strategic implication is that security is becoming an operating-model problem, not just a tooling problem. Microsoft frames the mismatch between weaponization and remediation as a widening window for attackers. 4 Rapid7 similarly notes that the pressing issue is not raw vulnerability count, but how quickly attackers exploit the vulnerabilities they choose. 3

For investors, that means the market should keep rewarding products that shorten response time, reduce manual triage, and help teams prioritize the exposures that actually move first. But it also argues for skepticism toward tools that promise complete visibility without showing how they compress the hours that matter.

What to do next

The evidence does not support a blanket claim that every patch window is collapsing. It does support a more useful one: the dangerous cases are getting much faster, and they are the ones operators cannot afford to miss.

If you run engineering or security, the question is not whether your average vulnerability response looks acceptable on a dashboard. It is whether your patching, detection, and prioritization loop can move fast enough for critical, internet-facing exposures when the attacker may already be ahead.

If it cannot, the gap between disclosure and exploitation is already large enough to matter.

Share this

Tags

Written by: 1 Minute Signal Editorial Team