Exposure Management Was Supposed to Fix a 30-Year-Old Problem. It Hasn’t.

Exposure management promised a shift from reactive patching to proactive risk reduction. It hasn't delivered: the average serious vulnerability still takes 60 to 100 days to fix. The reason is structural, built around activity instead of decisions.

Elad Ben Meir | Apr 28, 2026

Key takeaways

  • Exposure management is the continuous work of finding, prioritizing, and reducing risk across the attack surface, through to a verified fix. Most programs stop at the finding and the ranking.
  • Exploitation is now the most common way in. In the 2026 Verizon DBIR, vulnerability exploitation started 31% of breaches, ahead of credential abuse at 13%. For vulnerabilities in CISA's Known Exploited Vulnerabilities (KEV) catalog, the median time to full remediation rose to 43 days.
  • Mandiant's M-Trends 2026 puts mean time to exploit at an estimated -7 days. On average, exploitation now arrives before a patch does.
  • A defensible short list needs four answers, in order: reachable, exploitable, what it touches in the business, and who owns it. A short list is still not a smaller backlog.
  • Decision-Based Exposure Management changes the unit of work from the ticket to the decision. One decision resolves a class of exposure, present and future. Humans define judgment. Agents execute.

In late 2020, state-sponsored attackers breached a well-regarded network access control company. The way in was not a zero-day. It was a known vulnerability that sat in the backlog and never reached the top. Nobody owned it, and nothing drove it to closure. The company's co-founder, Ofer Amitai, went on to start Onit Security.

The lesson was not "patch faster." The system around the patch was broken.

What is exposure management?

Exposure management is the continuous practice of finding, prioritizing, and reducing risk across vulnerabilities, misconfigurations, identities, and attack paths, then verifying the fix. Vulnerability management finds and ranks CVEs. Exposure management is meant to finish the job.

Gartner introduced Continuous Threat Exposure Management (CTEM) in 2022 and gave the market a five-stage model: scoping, discovery, prioritization, validation, and mobilization. It pushed teams past point-in-time scanning toward continuous validation, and that was real progress. The fifth stage, mobilization, is where programs stall, because it is the one that depends on people acting.

The layers are easy to blur, so here is where each one stops.

LayerThe question it answersWhere it stops
Vulnerability managementWhich CVEs exist, and how severe are they?A ranked list of findings
Exposure assessment platformsWhat is exposed across assets, identities, and attack paths?A better picture, at a point in time
Exposure managementWhat matters here, who must act, and did it get fixed?Often stalls at mobilization
Decision-Based Exposure ManagementWhich decision resolves this whole class of exposure?Agents carry the approved decision to closure

Scanners and asset-inventory tools have made the picture rich and accurate. That is valuable. A picture still does not remediate anything.

Why does remediation still take so long?

Because finding is solved and fixing is not. The 2026 Verizon DBIR shows exploitation overtaking stolen credentials as the top way in, while the median time to fully remediate known exploited vulnerabilities rose to 43 days. More scanning does not fix a coordination problem.

Three 2026 reports describe the same gap from different angles. Each measures something different, so we cite each one as measured and do not blend them.

  • Verizon 2026 DBIR: vulnerability exploitation was the most common initial access vector, in 31% of breaches, ahead of credential abuse at 13%. For vulnerabilities in the CISA Known Exploited Vulnerabilities (KEV) catalog, the median time to full remediation rose to 43 days from 32, and only 26% were fully remediated, down from 38%.
  • Edgescan 2026 Vulnerability Statistics Report: the average time to remediate high and critical application and API vulnerabilities was 54.81 days in 2025.
  • Mandiant M-Trends 2026: mean time to exploit is an estimated -7 days, so exploitation routinely occurs before a patch is released.

Organizations have invested heavily in scanners, scoring engines, and AI-powered prioritization, and the queue still does not shrink. The cause is structural: the work is organized as tasks. A finding becomes a ticket, an owner hunt, and a review, even when the same pattern has appeared a hundred times. Every repeat costs as much as the first time.

What does a defensible decision need? Four answers, in order

Can an attacker reach it, does the attack work here, what does it touch that the business cares about, and who has authority to fix it. Only with all four can a decision be made once and kept.

1.  Reachable. Can an attacker get to this instance, and does the vulnerable code actually run? Those are two questions: a network path from a position an attacker plausibly holds, and whether the vulnerable function sits in the execute path rather than an imported dependency nobody calls.

2.  Exploitable. Given that they can reach it, does the attack work here, or has a compensating control already closed it? A flaw can be exploitable in a lab and unreachable in your network. Onit checks this read-only, through the EDR the customer already runs, in the customer's own environment.

3.  Business context. What does it reach that the business cares about: what it processes, what it connects to, what breaks if it stops, and which regulation covers the data on it. Severity scores and exploitability scores both miss this, because it exists only inside the customer's own network. The same flaw on a server in a PCI cardholder environment and on a sandbox rebuilt weekly is not the same finding.

4.  Ownership. Who has the authority to approve the fix? CMDB entries go stale, org charts trail how operations really run, and the person who set the system up left after an acquisition. In Onit's field observations, about half of remediation tickets bounce back to the security team because they reached the wrong owner, and about 30% stall for lack of context. "Who owns this?" can outlast the vulnerability window.

These four answers produce a defensible short list. A short list is not a smaller backlog. Filtering hides the backlog. Retiring whole classes of exposure shrinks it.

Next step (ungated, mid-content): See how one decision resolves a class of exposure on the Onit platform page: https://onit.security/platform/

How does Decision-Based Exposure Management work?

A person decides once for a class of exposure. Agents apply it to every current and future case: find the owner, plan the fix, open the ticket, chase it. It persists as an operating rule, so nothing is solved twice.

Decision-Based Exposure Management is Onit's operating model, and the category we are building. It changes the unit of progress from tasks to decisions, in five parts.

1.  The unit of work is the decision. Instead of one ticket per finding, an analyst makes a single judgment about a class of exposure: given this vulnerability profile, these asset characteristics, and this ownership structure, what should happen?

2.  One decision resolves thousands of exposures. It covers every matching case, present and future, across the environment. The scale is why this matters: Onit's largest deployment, a Fortune 500 enterprise running hybrid cloud and on-prem, carries more than 50 million active exposures across about 700,000 assets.

3.  The decision persists as an operating rule. It keeps applying on its own. Resolve. Don't re-solve.

4.  Ownership resolves from live context. Onit infers the owner from evidence in your CMDB, ticketing history, and Slack and Teams threads, rather than a static mapping.

5.  The knowledge compounds. The first decision covers one pattern. Each approved decision captures the team's judgment, so it survives team turnover, reorganizations, and acquisitions.

Think about what Cursor and GitHub Copilot did for a senior engineer: their judgment now applies at scale instead of being spent on repetitive work. Humans define judgment. Agents execute.

Where humans stay in control. Because the agents act, buyers ask four fair questions. What does the model do? Language models reason about the finding and infer ownership, and classic machine learning handles deduplication, clustering, and exploitability scoring. What data does it use? An organizational knowledge graph of assets, exploitability, and live-context ownership, and customer data never trains models. Where does human review sit? The system proposes and humans approve, as a permanent layer. Who authorizes disruptive actions? Humans do, with a full audit trail on every agent action and one-step rollback. You approve the decision once. Every later action runs under that rule, logged and reversible.

Why does machine speed change the answer?

Attackers now use AI agents to compress reconnaissance, exploitation, and handoff into minutes. If defense still runs at the pace of a weekly triage meeting, the exposure window is open before anyone has been assigned.

In November 2025, Anthropic reported that a state-sponsored group had used an AI agent to run a cyber-espionage campaign, with the AI performing 80 to 90% of the campaign. Mandiant's M-Trends 2026 found the median handoff from initial access to a secondary threat group had fallen to 22 seconds in 2025. In April 2026, Anthropic said its Claude Mythos Preview had found thousands of high-severity vulnerabilities, including some in every major operating system and web browser, in its own testing and with access restricted to vetted organizations.

If an attacker's agent can weaponize a vulnerability faster than your team can investigate the finding, the weekly queue has already lost. Machine-speed defense needs a closed loop: check reachability and exploitability, name the owner, propose a mitigation for approval, verify the fix. When the decision exists before the exploit does, remediation can close the gap with the attack. The goal is resolution in hours, not weeks.

What should security leaders change now?

Two shifts: from distributing tasks to capturing decisions, and from counting detections to measuring risk removed. Both start with letting people define judgment once, and letting agents execute it across the whole environment, under human approval.

Hiring more analysts to work through a bigger backlog has hit its ceiling. The attack surface grows faster than headcount.

First, move from task distribution to decision capture. Stop routing a task to a person for every finding. Capture the human decision about a class of findings, and let agents execute it.

Second, move from detection metrics to risk-reduction metrics. MTTR is a useful proxy. The harder question is how quickly you remove business risk, which means asking which exposures are dangerous in your environment, to your business processes, given your controls.

A program without a decision layer is a better map of the same stuck backlog. Decide once. Resolve forever.

End-of-post ask: See what Decision-Based Exposure Management looks like in your environment: book a demo.

Frequently asked questions

Exposure management is the continuous practice of identifying, prioritizing, and reducing an organization’s attack surface, including vulnerabilities, misconfigurations, and identity risks, with accountability for resolution. Unlike point-in-time scanning, it connects findings to business context, ownership, and a verified fix.

No. Scanners keep finding. Exposure management works from what they find: it adds reachability, exploitability, business context, and ownership, then drives the fix. Onit sits above the scanners a team already runs and replaces none of them.

Exposure assessment platforms scan, score, and report a picture of what is exposed at a point in time. Exposure management is the operational layer that assigns ownership, drives the fix, and verifies that risk fell. Assessment answers what is broken. Management answers what is being done about it and whether it worked.

Decision-Based Exposure Management is Onit’s operating model. A person makes one decision about a class of exposure, and agents apply it to every current and future case, finding the owner, planning the fix, and chasing it to closure. The decision persists as an operating rule, so the same exposure is never solved twice.

Because the cause is manual coordination, not missing tools. For every finding someone still has to check exploitability, find the owner, design a mitigation that does not break a dependency, and verify the fix. Faster tools speed up a manual pipeline without removing it. Encoding the decision removes the repeated steps.

Get a Demo