How every CVE gets scored, owned, and put on a clock.
The full technical walkthrough behind vulnerability management in Secuno: how versions are matched, how risks are scored and escalated, and how the workflow closes them.
Most CVEs will never be exploited. Secuno finds the ones that will.
A raw CVE list is a to-do list with no priority. Secuno scores every correlated vulnerability by exploit probability, flags the ones CISA lists as actively exploited, and checks whether working exploit code exists, so your team spends its hours on the few vulnerabilities attackers actually use, not the thousands they never will.
- 01EPSS probability, and how it moves
Every CVE carries its EPSS exploit-prediction score and percentile, with ninety days of history, so a low-CVSS bug under active exploitation still rises to the top. When a score jumps past your alert threshold (50% by default) on a CVE a client has an open risk for, that client's administrators hear about it the same day.
- 02CISA KEV, with context
Known Exploited Vulnerabilities are flagged with their remediation due date and ransomware-campaign linkage, not just a boolean.
- 03Exploit availability
Public exploit signals (ExploitDB, Metasploit module, proof-of-concept on GitHub) sharpen the priority beyond the base score.
- 04Nobody exploiting it? It scores lower.
A CVE with no KEV listing and no public exploit is discounted by its EPSS: under 1% it keeps 85% of its score, between 1% and 10% it keeps 95%. Never by more than one severity band, never while EPSS is unknown, and never in a CVE's first five days. A CVSS 9.6 that nobody exploits lands at 82, High instead of Critical, so the Critical queue holds what attackers actually use.
- 05Is the vulnerable code running?
Every open finding records whether the vulnerable code is loaded in memory, running, used in the last 30 days, untouched for 90 days or more, or a bundled library no process has loaded. Code that is actually running keeps its full weight, so it is fixed first. Each risk shows one line of evidence: which process runs it, and from where. A device the agent could not measure is never scored down, and a known-exploited CVE stays at least High whatever its runtime state.
From an installed version to an owned risk, step by step.
Automatic and idempotent. Every version on every device runs the same path, and the same input never produces a duplicate risk.
- INVENTORY
Read every installed version
The Go agent reads installed applications from three registry hives (x64, x86, per-user), plus Windows KB patches and Winget packages, with publisher, version and install date. On Linux it reads the distribution's own package list. Collected every hour, with a full sweep every night at 03:00.
- INDEX
Match without scanning the world
An inverted product-to-CVE index resolves matches in near-constant time instead of scanning the full catalogue, with version-constraint parsing across semver, ranges, exact versions and ecosystem-specific formats.
- CORRELATE
CPE-first, and honest about the tail
Exact CPE-to-CPE matching is the primary path, and Windows itself is matched against Windows CVEs by its full build number, so a host one update behind matches and a fully patched one does not. Findings that an operating system update fixes, on Windows or in a distribution's own packages, get their own category, OS Vulnerabilities, apart from the ones fixed per application. Linux packages are checked against the distribution's own advisories from OSV instead of NVD, using that package manager's own version ordering (dpkg, rpm, apk), so a backported fix counts as fixed. Private AI resolves the fuzzy tail, and every finding is then graded on how certain the match is: confident enough is accepted, borderline is shown and flagged for a human, and too uncertain is dropped to informational and kept out of the default queue. A CVE known to be actively exploited is never hidden.
- ENRICH
EPSS, KEV and exploit intel
Each CVE record carries its CVSS score on the newest scale NVD has for it (4.0 first, then 3.x, then 2.0), always labelled with its version, plus EPSS score and percentile, CISA KEV status, and exploit-availability flags. New CVEs arrive from CVE.org every hour. NVD enrichment (CVSS and CPE rules, read from a partner-maintained NVD database rather than the NVD API), Linux distribution advisories from OSV and CISA KEV refresh every six hours, and the GitHub Advisory Database every four hours. GitHub advisories describe language packages, so they are not matched to device software: together with OSV they check the front-end libraries a client's website loads.
- SCORE
Deterministic first, AI on a leash
A 0 to 100 score is computed from things that can be checked: the severity band the source declared, how strong the evidence is, whether an attacker can reach the affected path, whether your controls block that specific path, and whether the vulnerable code is actually running. Private AI adds a plain-English impact summary, remediation steps and a bounded adjustment that can never outweigh the measured controls.
- OWN
One risk, with an owner and a clock
A deduplicated risk lands with an assignee, an SLA due date that follows its earliest open finding, MTTD and MTTR tracking, evidence attachments and a full audit trail. A due date a person typed is never moved. After each inventory it re-verifies, and reopens automatically if the vulnerable version returns.
Every correlation finding carries its own match confidence, and it is on the risk where an analyst can see it. This one scored 95, well clear of the bar for automatic acceptance. Between 40 and 59 a finding is shown but flagged for a person; below 40 it drops to informational and leaves the default queue, without being deleted.
Scoring runs on SECAI, on our own infrastructure in the Netherlands. See the full security and infrastructure model
A new exploit goes public. Every affected risk escalates before your analysts log in.
CISA adds a CVE to the Known Exploited Vulnerabilities catalogue. Secuno does not wait for the next sweep: it correlates that CVE across every client device at once, escalates each affected risk and pulls its due date in. No one had to be watching the feed.
- CISA KEV synchronised every six hours; a newly listed CVE triggers a targeted correlation at once
- Affected clients get a critical in-app notification and an email
- Each affected risk is rescored and its due date pulled in to the new severity's SLA, never later
- The fixing update moves to the Emergency patch lane, with the remediation deadline and ransomware linkage on the risk
Correlation is only as good as the inventory under it.
The Go agent delivers 20+ data types per device on an hourly inventory, a nightly full sweep and a five-minute heartbeat, signed with Ed25519 and sent over TLS 1.3 to our infrastructure in the Netherlands. Every field is a place a risk rule can fire.
| KB ARTICLE | TITLE | STATE | APPLIED |
|---|---|---|---|
| KB5066128 | Security intelligence update for Microsoft Defender Antivirus (version 1.457.211.0), current channel (broad) | missing | — |
| KB5082417 | 2026-04 Cumulative update for Windows 11 Version 23H2 for x64-based systems | installed | 4/16/2026 |
| KB5054156 | Servicing stack update for Windows 11 Version 23H2 for x64-based systems | installed | 4/09/2026 |
| KB5049209 | .NET 8.0 cumulative update for Windows 11 Version 23H2 | installed | 3/28/2026 |
This interface is for illustrative purposes only, representative of the actual application. It is not a screenshot.
A full risk workflow, not just a finding.
A correlated CVE is where the work starts. Secuno wraps every risk in the workflow an auditor and an analyst both expect.
Owned, on SLA, and in a queue
Each risk has an assignee and a due date. The list opens on the queues a team works from: My risks, Unassigned, Overdue, Awaiting approval and Watched, so nothing quietly ages past its deadline.
Status with proof
Open, In Progress, Mitigated, Accepted or False Positive. Accepting a risk always takes two people: one proposes it with a justification and any compensating controls, and someone else approves it, with an expiry of at most a year, after which the risk reopens automatically. There is no admin exception and no way to set a risk to Accepted directly. A deleted risk can be restored for 30 days, with its history, notes and evidence.
Auto-verified remediation
After each inventory the platform rechecks closed risks. If the vulnerable software is back, the risk reopens itself with a note and a notification to the owner.
Proof the fix happened
Every device and CVE pair is tracked from the day it opened to the day it closed, and the reason it closed is read from inventory: updated to a named version, removed, no longer affected, device retired, or no longer detected. Only an update or a removal counts as a fix. Fixed per month, time to fix and SLA met by severity are on one page, with a CSV export, and the same history shows on the device, the CVE and the risk.
Deduplicated and audited
A correlation key stops duplicate risks at the source. @mentions, watch subscriptions, evidence attachments and a chronological timeline cover the audit trail.
Bulk at fleet scale
Bulk status change, assignment, watch and merge, plus multi-select CSV export and a PDF of the risk list, built for large remediation sprints across hundreds of clients.
The fix is on the risk
A Windows risk shows the update (KB) that fixes it on that client's devices, joined through Microsoft's own security-update data, and the per-device winget upgrade with a one-click Remediate that acts on that device only. The patch and the vulnerability are one conversation, not two pages.
Grouped by the fix that closes them
Group the list by Fix and the open vulnerability risks turn into the actions that close them: a winget upgrade or a Windows update the client's devices are missing. Each fix shows its devices, the risks it closes outright, the ones it only reduces, and the share of the client's open risk it removes. Remediate all allows the upgrade once on every device that still needs it. With one client in scope, the same view offers the Ranked plan: every open finding on the client, turned into the fixes that remove the most risk first.
Analyst to board, in one click.
Every view shows exactly what its reader needs, from a single CVE to a portfolio-wide security posture.
- Reports for every reader: Risk Summary, Vulnerability, Compliance, Executive Summary, Period Comparison, Microsoft 365 Posture, Portfolio, Board report, QBR and Custom
- Export to per-client branded PDF, CSV or JSON
- Per-client branding: logo, primary colour and footer text
- Scheduled delivery with timezone support. Risk and executive summaries open with a plain-language narrative built from the period's attack stories, in Dutch with English beneath

How the engine works, answered plainly.
What is CVE correlation?
It is the process of matching the exact software versions installed on a device against a database of known vulnerabilities (CVEs), so that instead of a generic advisory feed you get a precise list of which of your devices are actually affected by which vulnerability. Secuno does this automatically for every managed device, several times a day.
How does Secuno decide which vulnerabilities matter?
Every correlated CVE is scored by its EPSS exploit probability, flagged if CISA lists it as actively exploited (KEV), and checked for public exploit availability. On top of that, Secuno works out whether an attacker can actually reach the affected path on that specific device, and whether your existing controls block that path. A critical CVE in a service bound to loopback is not the same finding as the same CVE on an internet-facing host, and the queue reflects that. Last, it checks whether the vulnerable code is actually running: a program nobody has opened in 90 days ranks a little lower than one loaded in memory right now.
Does my data go to an external AI provider?
No. Scoring runs on SECAI, Secuno's own inference engine, on our infrastructure in the Netherlands. No CVE data, device inventory or risk detail is sent to a third-party model provider.
How quickly is a newly exploited vulnerability reflected?
New CVEs arrive from CVE.org every hour, and NVD enrichment and the CISA KEV catalogue refresh every six hours. A CVE newly added to KEV does not wait for the next sweep: it triggers a targeted correlation at once, affected clients get a critical notification and an email, each affected risk escalates with its due date pulled in, and the update that fixes it moves to the Emergency patch lane.
What stops the queue filling with false positives?
Every correlation finding is graded on how certain the match is, and the cuts were calibrated against a full production set of findings rather than picked by feel. They keep calibrating: every week the engine fits them to the Correct match and Wrong match verdicts your analysts record and, once it has at least 50 of each, proposes moving a cut by at most five points. An administrator applies the proposal; nothing moves on its own. Confident matches are accepted automatically, borderline ones are shown and flagged for a human decision, and matches we do not believe are dropped to informational and kept out of the default queue. They are not deleted and stay retrievable. A version the engine cannot read with confidence, such as a packaging tag glued to a version number, ends the comparison and keeps the CVE out rather than asserting a match. A version read from a website's banner rather than from an inventory is weaker evidence, so it is capped at the review tier and never accepted automatically. One deliberate exception: a CVE listed as actively exploited or with a high EPSS score is never hidden, at worst it lands in the review queue. A suppressed false positive costs an analyst nothing; a suppressed known-exploited CVE is the one mistake you cannot recover from.
Does it work on Linux without flagging patched hosts?
Yes. The classic Linux false positive comes from distributions backporting a security fix without changing the upstream version number, so a package matched against NVD looks vulnerable on a fully patched host. Secuno does not match Linux packages against NVD. It reads the distributions' own advisories through OSV (Ubuntu, Debian, Alpine, Rocky Linux, AlmaLinux and openSUSE) for the packages your fleet actually runs, and compares versions the way that distribution's package manager does (dpkg, rpm or apk). A host one update behind matches; the same host after the update does not.
Can we tell the score something the agent cannot measure?
Yes, through device groups. A group can carry adjustments from a fixed list of reasons: not reachable over the network, on an isolated or test segment, no sensitive data, being decommissioned, internet facing, sensitive or regulated data, business critical, legacy or cannot be patched, shared or kiosk device, or a reason in your own words. Each adjustment is bounded, a reason counts once however many groups repeat it, and the rule that context can move a finding at most one severity band still applies first, so a group can reorder findings but never turn an informational note into a critical one. The risk page lists these adjustments separately from the measured controls, because they are an operator's statement rather than a measurement. A group can also declare that membership is itself a finding, for example devices on an unsupported operating system.
What about libraries bundled inside applications?
They have their own category, Bundled Vulnerabilities. A CVE in a copy of OpenSSL, zlib or libwebp that an application ships in its own folder is fixed by updating that application, not the library, so it is listed apart from installs of the library itself. A bundled copy that no running process has loaded is marked as on disk but not loaded, and ranks slightly lower than one in use.
Do findings without a CVE get the same treatment?
Yes, and this changed recently. Findings from custom rules, Microsoft 365 posture, agent change events and manual entry are the majority of what an operator actually opens, and they used to arrive with no score and no explanation because both required a CVE. The source now declares a severity band, scoring ranks within that band, and the same plain-English explanation panel is filled in for them. Findings that predate this are backfilled rather than left blank.
Turn 400,000 CVEs into a queue your team can actually clear.
A 30-minute walkthrough, from agent rollout to the first scored risk and the board report. Bring your hardest client environment.