Exploitation now beats patching. Your 30-day SLA is fiction.

Exploitation now lands in days while median remediation runs weeks. Why the mean time to exploit vs. MTTR gap makes the 30-day patch SLA obsolete.

Onit Security | Sep 28, 2026

TL;DR

  • Exploitation now beats patching. Attackers weaponize disclosed flaws in days; median remediation still takes weeks.
  • Mean time to exploit for KEV-listed edge devices is zero days. Mean time to remediate the same class is 43 days.
  • Sixty percent of breaches involved a vulnerability with a patch already available.
  • The 30-day SLA measures ticket compliance, not risk reduction, and was never built for a threat that arrives same-day.
  • Decide once. Resolve forever. One decision closes a whole class of exposure instead of one ticket at a time.

It is Friday afternoon. A vendor publishes an advisory for a critical vulnerability in a product you run on a few hundred internet-facing devices. Proof-of-concept code is on GitHub by the evening. Over the weekend, mass scanning finds your edge before your on-call engineer has finished reading the write-up. By the time your team triages the finding on Monday morning, the vulnerability has already been exploited in the wild for two days.

Your remediation SLA gives them 30 days to close it.

Two clocks are running, and they are not the same clock. One is the attacker's mean time to exploit, the gap between a vulnerability going public and it being used against you, now measured in hours. The other is your mean time to remediate, or MTTR, still measured in weeks. That gap is not a detail. It is the whole game, and attackers learned to live inside it long before most programs admitted it existed.

For most of this field's history, MTTR ran comfortably ahead of mean time to exploit. That order has now reversed, and this post is about what the reversal means.

The exploitation window did not shrink. It collapsed.

For most of the history of this field, defenders had time. A CVE would publish, a working exploit would take weeks or months to appear, and a 30-day patch cycle sat comfortably ahead of the threat. That world is gone, and it did not fade out gradually. It fell off a cliff.

Vulnerability exploitation is now the single most common way attackers break into enterprises, responsible for 31% of breaches in the 2026 Verizon DBIR. It overtook stolen credentials for the first time in the report's history, climbing from 20% the year before while credential abuse as an entry point fell to 13%. And for the vulnerabilities that matter most, exploitation now begins almost immediately. The median time from public disclosure to mass exploitation is five days for vulnerabilities on CISA's Known Exploited Vulnerabilities list, and zero days for the internet-facing edge devices, the VPNs and firewalls, that attackers hit first (2026 Verizon DBIR).

Zero days. Not a zero-day in the classic sense, but a known, published flaw being weaponized on the same day it becomes public.

Three forces drove this. Exploit development is increasingly automated, so the distance between an advisory and working attack code keeps shrinking. Ransomware crews and access brokers run continuous, internet-wide scanning, so an exposed asset is found in minutes rather than weeks. And edge devices, the VPNs, firewalls, and file-transfer appliances that sit at the perimeter, have become the favored target precisely because they face the internet, rarely get patched on time, and hand the attacker a direct route inside. The adversary does not need to be sophisticated anymore. They need to be fast, and the tooling makes them fast by default.

The exploit does not wait for your maintenance window. It shows up first.

Your remediation still runs on a calendar

Now put the defender's clock next to the attacker's, and be honest about what the numbers actually say.

The most optimistic figure in the industry is a median of 43 days to remediate a vulnerability that already sits on CISA's Known Exploited Vulnerabilities list (2026 Verizon DBIR). It is worth sitting with that number, because it is the good one. KEV-listed flaws are the known-dangerous subset, the ones a team should be driving hardest to close. And even that number is getting worse. It was 32 days the year before. The lane everyone points to as evidence of progress is slowing down, not speeding up.

Step outside the fast lane and the picture darkens. Across all discovered flaws, half are still open after 243 days (Veracode, 2026). That is a half-life, not an average, which means the other half take even longer, and a long tail is never fixed at all. Broader MTTR benchmarks that blend severities land somewhere between 63 and 270 days depending on the source and the sector.

Here is the point. Every one of those numbers is measured in weeks or months. The attacker's number is measured in seconds. You do not need a chart to feel the mismatch. A process that closes its most urgent work in 43 days cannot defend against a threat that arrives in zero. They are not in the same order of magnitude, and no amount of color-coding on a dashboard changes the arithmetic.

Mean time to exploit vs. mean time to remediate

Put the two side by side and the problem states itself. Mean time to exploit, for the vulnerabilities attackers actually pursue, is now close to zero. Mean time to remediate is still measured in weeks. MTTR was a useful number when it ran ahead of the threat. It now runs an order of magnitude behind, and reporting it on its own hides the gap instead of surfacing it. The real question is not how fast you patch. It is whether your mean time to remediate is shorter than the attacker's mean time to exploit. For most organizations today, it is not, and no service-level agreement rewrites that arithmetic.

Where the 30-day SLA came from, and why it stopped working

The 30-day SLA did not appear out of nowhere. It came from compliance. Audit regimes and security frameworks needed a measurable, defensible line, and "close critical vulnerabilities within 30 days" was clean, auditable, and roughly aligned with how fast attackers moved at the time. For years it was a fair proxy. It gave security teams a target and gave auditors a box to check.

The trouble is that the metric measured the wrong thing, and no one noticed until the threat outran it. Patch SLA compliance tracks whether a ticket closed inside a window. It says nothing about whether risk actually went down, or when, or by how much. It treats every critical the same, whether it sits on an isolated test box or on a domain controller facing the internet. It rewards closing easy tickets quickly and quietly tolerates the hard ones aging out in the backlog.

And it assumes a patch is the answer. Often it is not. Sometimes there is no patch yet. Sometimes the patch breaks a production dependency and cannot ship on the vendor's schedule. Sometimes the system is an unsupported appliance or an operational-technology device that cannot be taken offline. In every one of those cases, a 30-day patch SLA is meaningless. The clock ran out on a fix that was never going to happen inside the window, and the metric had no way to say so.

The proof is in the breaches. Sixty percent of them involved a vulnerability that already had a patch available and was never applied (Ponemon Institute / ServiceNow). The fix existed. The window closed anyway. A deadline that expires after the attacker is already inside is not a deadline. It is paperwork.

One known flaw. One state-sponsored breach.

Onit exists because of exactly this failure. Before starting the company, our co-founder Ofer Amitai watched state-sponsored attackers walk into his organization through a single known vulnerability. Not an exotic zero-day. A published flaw, with a fix available, buried in a backlog no one could reach the bottom of. It was never patched, because no one owned it and nothing drove it to closure. The window stayed open long enough for a determined adversary to find it and use it.

That is what the exploit-versus-patch gap looks like from the inside. It is not a statistic on a slide. It is one flaw, in one backlog, that everyone technically knew about and no one closed in time.

Patching speed is the wrong race

So the instinct is to patch faster. Tighten the SLA to 14 days, then seven. Automate the patch pipeline. Run more remediation sprints. All of it helps at the margins, and none of it wins the race, because the race itself is defined wrong.

When the exploit arrives before the patch, patching faster cannot save you. The question was never how quickly you can deploy a vendor fix. It is how quickly you reduce the risk, by whatever means the situation allows. That might be a patch. It might be a compensating control, a firewall rule, a segmentation change, a configuration hardening, a validated mitigation that holds the line until a real fix ships. The attacker does not care which lever you pulled. They care only whether the door was still open when they arrived.

This is why the industry has to stop talking about remediation and start talking about resolution. Remediation is one path to safety. Resolution is safety itself. And the metric that follows from it is not SLA compliance. It is time to risk reduction: how fast did the real exposure go down, and can you prove it. Almost no one measures that today. The teams that start will find it is a very different number, and a far more honest one, than the percentage of tickets closed inside a 30-day window.

You cannot out-hire the math

Suppose you accept all of this and decide to simply move faster. Hire more analysts. Run the program harder. Here is what stops you: volume.

Known CVEs are projected to pass one million by 2030, up from roughly 277,000 in 2025 (Gartner). Multiply that by the number of assets in a large enterprise, hybrid cloud plus on-prem plus edge, and the count of individual exposure instances runs into the tens of millions. One large environment we work with carries more than 50 million active exposures across roughly 700,000 assets. No headcount plan touches a number like that. A team that doubles in size still processes findings one at a time, at human speed, while exploitation happens across thousands of assets at once, at machine speed.

The gap is not an effort problem. It is a structure problem. The unit of work is wrong. As long as every finding generates its own investigation, its own ownership hunt, its own ticket, and its own validation, the work grows in a straight line against a threat that grows exponentially. You lose that race by definition, no matter how hard anyone tries.

The only way to win is to stop doing the work one finding at a time.

Decision-Based Exposure Management: the unit of work becomes the decision

That model has a name: Decision-Based Exposure Management. It begins with a plain observation. Most of the effort in a remediation program is not the fix itself. It is the repetition: the same investigation, the same ownership hunt, the same triage and validation, run again for every instance of a problem the team has already seen a hundred times.

Decision-Based Exposure Management removes the repetition. Instead of turning every finding into its own ticket, it groups related exposures into a single decision: what the exposure is, which assets it affects, who owns it, and what resolution looks like given your environment. A human makes that judgment call once. Agents then execute it across every matching asset at the same time, and the decision persists as an operating rule, so the next time the same pattern appears it is handled without anyone starting over. Humans define judgment. Agents execute.

Take a concrete case. An outdated encryption library turns up on six hundred servers, same configuration, same owning team. In a task-based program that is six hundred investigations, six hundred ownership lookups, six hundred tickets, and six hundred validations. In a decision-based one it is a single decision: this library, these six hundred assets, this owner, this fix, or this compensating control if the fix cannot ship. A person reviews and approves it once. Agents apply it across all six hundred, and the next server that appears with the same library is handled by the same rule, without a new ticket.

Speed like that only counts if it is safe, so the human stays in the loop where it matters. The system proposes, a person approves anything that could cause an outage, and every action can be rolled back in one step. After execution, the decision is checked against whether the exposure actually dropped, not whether a ticket closed. The judgment is human. The repetition is not.

This is a different thing from making the old process faster. Better prioritization and smarter ticketing still handle findings one at a time. They just make the tasks quicker. Decision-Based Exposure Management removes the need to repeat them. One decision resolves thousands of exposures, the ones you have today and the ones that appear tomorrow, and the organizational knowledge inside that decision compounds instead of resetting with every new finding. Decide once. Resolve forever.

The window is measured in hours. Your response has to be too.

The 30-day SLA was an honest answer to a question that no longer exists. Attackers do not take 30 days. They take minutes, and they have the tooling to do it across every exposed asset you own the moment a flaw goes public.

You will not close that gap by patching faster or by hiring more people. You close it by measuring the right thing, time to risk reduction, and by resolving exposure in decisions instead of tickets. That is the difference between hours and weeks: one approval resolves a whole class of exposure everywhere it exists, instead of a thousand tickets inching through a queue while the window stays open.

Decide once. Resolve forever. The alternative is to keep reporting SLA compliance to the board while the window stays wide open, which is a fiction most of the room has already stopped believing.

See how Onit closes the exposure window in hours instead of weeks. Get a demo.

Frequently asked questions

Mean time to exploit is the time between a vulnerability becoming public and attackers using it against real targets. For the vulnerabilities that matter most it has collapsed toward zero. The 2026 Verizon DBIR puts the median at zero days for internet-facing edge devices and five days for all vulnerabilities on CISA’s Known Exploited Vulnerabilities list.

Mean time to exploit measures how fast the attacker moves. Mean time to remediate, or MTTR, measures how fast you respond. MTTR was a sound metric while it ran ahead of exploitation. Today the median time to remediate a known-exploited vulnerability is 43 days (2026 Verizon DBIR), while exploitation often begins in days or hours. When mean time to exploit is shorter than mean time to remediate, the exposure window stays open by default.

No. A 30-day SLA was built for a threat that took roughly 30 days to arrive, and that world is gone. Sixty percent of breaches involve a vulnerability that had a patch available and was never applied (Ponemon Institute / ServiceNow), and exploitation now often precedes the patch entirely. A fixed 30-day window measures ticket compliance, not risk reduction.

Measure time to risk reduction rather than patch speed: how fast the real exposure drops, whether through a patch or a compensating control. Then change the unit of work from the individual ticket to the decision, so one human judgment resolves a whole class of exposure across every affected asset at once. This is the basis of Decision-Based Exposure Management.

Get a Demo