Grype vs Trivy vs OSV-Scanner (and Patch Proof) on the same two Linux servers

Updated November 15, 2026By HostCrust19 min read

Both servers were fully patched. On Debian 12, Trivy reported 634 CVEs, OSV-Scanner 653, Patch Proof 2,643 and Grype 8,126. On AlmaLinux 9.8, Grype reported 12,598, Trivy 795, and OSV-Scanner and Patch Proof 0.

For Debian's installed user-space packages, all four tools agreed on 630 CVEs, none of which had a Debian fix yet. Almost all of the remaining gap comes from two things:

  • Which kernel each tool checks, and how it reads the kernel's version.
  • Whose vulnerability data it trusts. On AlmaLinux, Grype used Red Hat's data, which lists 7,198 of these CVEs as unfixed or won't-fix. AlmaLinux's own data, which Trivy and Patch Proof use, lists only released fixes.

How many CVEs did each scanner report?

We ran four free scanners on two real servers within 40 minutes on 14 Nov 2026, from 18:41 to 19:20 UTC. Both servers were fully patched: apt and dnf had nothing left to install. The counts below are unique CVE IDs, each counted once however many packages it was reported against.

CVEs reported by Trivy, Grype, OSV-Scanner and Patch Proof on the same two patched servers Unique CVE IDs per tool. Debian 12: Trivy 634, Grype 8,126 (630 installed packages, 6,112 kernel image files, 1,384 only from cached .deb files), OSV-Scanner 653, Patch Proof 2,643 (649 packages, 1,994 running kernel). AlmaLinux 9.8: Trivy 795 (all on an old kernel package), Grype 12,598 (559 user-space packages, 12,039 kernel), OSV-Scanner 0 (no package sources found), Patch Proof 0. Debian 12.15 (web-01) Trivy 0.76.2 634 Grype 0.121.3 8,126 OSV-Scanner 2.7.1 653 Patch Proof 1.0.0 2,643 AlmaLinux 9.8 (web-02) Trivy 0.76.2 795 Grype 0.121.3 12,598 OSV-Scanner 2.7.1 0 (no package sources found) Patch Proof 1.0.0 0 0 2,500 5,000 7,500 10,000 12,500 Unique CVE IDs reported User-space packages Kernel (packages or image files) Cached .deb files only
Unique CVE IDs each tool reported on 14 Nov 2026, split by where the finding came from. Not shown: 43 (Trivy) and 20 (Grype) CVEs in Go binaries that another test had put on the Debian VM.

Debian 12.15 (web-01)

372 packages are installed. The running kernel is 6.1.190-1. The older 6.1.128-1 kernel is still installed.

Tool and modeCVEs reportedInstalled user-space packagesKernelFrom other filesWith a fix available (per the tool)
Trivy 0.76.2, trivy rootfs /6346340None counted (43 more in Go binaries from another test)0
Grype 0.121.3, grype dir:/8,1266306,112, from the two kernel image files in /boot, matched through NVD1,384 found only in .deb files in the apt cache (20 more in Go binaries)0 for installed packages
OSV-Scanner 2.7.1, scan source -r653653, of which 234 are flagged “unimportant” by Debian0None0
Patch Proof 1.0.02,6436491,994, for the running kernelNone0

All four agree on 630 user-space CVEs. Grype's 630 are all in Trivy's 634, and Trivy's are all in OSV-Scanner's 653. Every one of them has no Debian fix: Trivy marks them affected (474), fix_deferred (149) or will_not_fix (11), and Grype marks them not-fixed or wont-fix. Debian's tracker gives the 630 these statuses:

  • 228 “unimportant”
  • 145 <postponed>
  • 76 <no-dsa>
  • 10 <ignored>
  • 171 open with no decision yet

Installing updates won't clear any of them.

In its JSON, OSV-Scanner tags 234 of its CVEs as Debian “unimportant”. Its documentation shows such entries moved out of the main table, into a “Filtered Vulnerabilities” list.

AlmaLinux 9.8 (web-02)

347 packages are installed and the running kernel is 5.14.0-687.66.2.el9_8. The older kernel-core 5.14.0-570.12.1.el9_6 is still installed, and so is a dracut rescue kernel from the original AlmaLinux 9.1 install (December 2022), which has its own GRUB entry.

Tool and modeCVEs reportedInstalled user-space packagesKernelWith a fix available (per the tool)
Trivy 0.76.2, trivy rootfs /7950795, all on the old kernel-core package that isn't running795
Grype 0.121.3, grype dir:/12,598559: 487 affected with no fix, 25 won't fix, 47 fixed only in RHEL 9.9 builds7,602 on the installed kernel packages (6,717 on the running one), plus 4,437 found only through the rescue kernel image963: 916 on the kernel packages, mostly the old one; 47 in user space, with fixes only in RHEL 9.9 builds
OSV-Scanner 2.7.1, scan source -r0“No package sources found”: it didn't read the RPM database
Patch Proof 1.0.0000 (running kernel checked)0. It listed 1,530 CVEs as fixed by backport and 351 as fixed.

Grype needed more memory than this 1.7 GB server had. The kernel's OOM killer stopped it twice, the first time at 1.4 GB of memory. It finished once we added a temporary 3 GB swap file, which we removed afterwards.

Why do Trivy, Grype and OSV-Scanner report different numbers?

For every CVE that not all four tools reported, we checked the distribution's own data. On Debian we used the security tracker, for every CVE. On AlmaLinux we used each tool's data source, plus Red Hat's and AlmaLinux's records for a sample. Each CVE got one reason:

ReasonDebian 12AlmaLinux 9.8Which tools it affects
Kernel image files matched by version string against NVD ranges5,4044,436Grype
Red Hat lists the CVE as unfixed or “will not fix”; AlmaLinux's data has no entry–7,198Only Grype reports these
Running kernel has no Debian fix, but the kernel packages aren't matched1,994–Only Patch Proof reports these
Only the old, installed-but-not-running kernel package is affected(inside the 5,404 above)887Trivy and Grype report them; Patch Proof doesn't
Found only in .deb files in the apt cache108–Grype
A newer fix exists for RHEL 9.9 but not yet for AlmaLinux 9.8–78Grype
Database timing, and Debian triage states23–Mixed

1. The kernel: which one, and how its version is read

The kernel causes most of the disagreement, in three different ways.

  • Grype matched kernel image files, not just packages. On Debian it catalogued /boot/vmlinuz-6.1.0-54-amd64 and /boot/vmlinuz-6.1.0-31-amd64 as binaries. It read their versions as 6.1.0-54-amd64 and 6.1.0-31-amd64, Debian's kernel ABI names, and matched them against NVD's upstream ranges. Both files got the same 6,112 CVEs. For 4,706 of them, the suggested fix is a 6.1.x release no newer than the 6.1.190 that's running. Debian's tracker shows how those 6,112 break down:
    • 2,818 are fixed in both installed kernels.
    • 2,444 are fixed in the running kernel but still affect the old one.
    • 136 never affected Debian 12.
    • 708 are really open.
    • 6 aren't in the tracker under linux.
    On AlmaLinux, Grype matched only the rescue kernel image this way. The rescue kernel really is old (5.14.0-162.6.1.el9_1). AlmaLinux fixed 680 of the matched CVEs after that build, but 26 before it, and 6,399 have no AlmaLinux kernel advisory at all.
  • Trivy, OSV-Scanner and Grype's package matcher found nothing for Debian's kernel. Debian 12's linux-image-* packages are built from the source package linux-signed-amd64, while Debian's tracker files kernel CVEs under linux. Trivy looks advisories up by source name (see its Debian detector), so it finds none. Patch Proof maps the running kernel to linux. It reported 1,994 CVEs with no Debian fix for 6.1.190-1, and Debian's tracker, about five hours later, still had 1,984 of them open. Of the 1,994:
    • 1,876 have no urgency assigned yet.
    • 62 are “unimportant”.
    • 41 are <ignored>.
    • 4 are <postponed>.
    • The other 11 are plain no-DSA, “undetermined”, or no longer listed as open.
  • Running versus installed. Patch Proof judges only the running kernel. Trivy and Grype judge every installed kernel package they can match. On AlmaLinux, Trivy's 795 CVEs are all on the old kernel-core 5.14.0-570.12.1.el9_6, and every one is fixed in the running kernel. Whether that counts depends on you: an old kernel that's still installed can be booted from GRUB. Removing it makes the finding go away; leaving it as a fallback keeps it.

To see what's actually running, compare uname -r with the newest installed kernel. Our guide to checking the kernel version shows how.

2. Unfixed CVEs: whose data you trust

Grype detected AlmaLinux 9.8, then matched the installed RPM packages against Red Hat's data: all 43,205 of its RPM matches came from the redhat:distro:redhat:9 namespace, which links to Red Hat's CVE pages. Anchore's OS support table lists the AlmaLinux OSV database as the AlmaLinux source, but every match in this run cited Red Hat's data. Red Hat publishes a status for CVEs it hasn't fixed:

  • “Affected”
  • “Fix deferred”
  • “Will not fix”

So Grype listed 6,686 such CVEs on the running kernel and 512 in user space. AlmaLinux's errata contain released fixes only, and Trivy (its docs say unfixed vulnerabilities aren't supported for AlmaLinux) and Patch Proof (its README says the same about AlmaLinux data) use them. Neither reported any of those 7,198.

Red Hat's data also lists 78 fixes in RHEL 9.9 builds, such as openssl-3.5.9-1.el9_9 and krb5-1.21.1-10.el9_9.2, that the AlmaLinux 9.8 repositories didn't carry yet. Grype reports these as fixable; the server can't install them.

On Debian, all four tools read the same source, Debian's tracker, so they nearly agreed on user space.

3. Files that aren't installed packages

grype dir:/ catalogues every package it can recognize on the disk, not only those in the package database:

  • The apt cache. It counted the .deb files in /var/cache/apt/archives: 2,625 CVEs in all, 1,384 of which appear nowhere else in its results. Only 108 of those were reported by no tool for an installed package; most of the rest are Debian kernel CVEs, matched through a cached kernel build-tools package, that Patch Proof also lists for the running kernel. Some of those files had been there since the provider built the image (openssh-server_1:9.2p1-2, curl_7.88.1-10). Others came from that day's patch run and from other tests on the VM.
  • Language packages. Trivy and Grype also scanned Go binaries that another test had placed under /root. We left those 43 and 20 CVEs out of the comparison.

Running sudo apt-get clean, or adding --exclude './var/cache/apt/archives/**', removes the apt-cache findings.

4. Small differences: timing and triage

The user-space disagreements on Debian came to just 23 CVEs:

  • Database timing. Each tool's data was built at a different time: Grype 06:19 UTC, Trivy 13:05 UTC, Patch Proof 14:03 UTC, and OSV-Scanner live at 18:44 UTC. CVEs published or re-triaged in between appear in some tools and not others.
  • “Undetermined”. Debian hasn't triaged five CVEs yet. OSV-Scanner and Patch Proof listed them; Trivy and Grype didn't.
  • Stale data. Three CVEs were still listed against Debian 12 in OSV.dev's copy of Debian's data, but not in Debian's tracker: two have no package entry any more, and one is now marked not affected. OSV-Scanner and Patch Proof reported them.
  • <ignored>. Debian ignores one CVE, in zlib, because no binary package is affected. Grype left it out; the other three reported it.

The samples below show each kind.

The sampled CVEs, checked against the trackers

We sampled 18 disagreements on Debian and 16 on AlmaLinux and looked each one up the same day. For Debian we used the security tracker. For AlmaLinux we used Red Hat's CVE data and AlmaLinux's errata, plus an OSV.dev query for the installed AlmaLinux build. Each CVE links to the record we checked.

Debian 12 (web-01)

CVEReported byWhat the distribution’s data saysWhy the tools differ
CVE-2026-99286Patch Prooflinux: 6.1.190-1 (bookworm-security) vulnerableRunning kernel, no Debian fix. Trivy, OSV-Scanner and Grype's package matcher never link linux-image-* to Debian's linux records. Grype's kernel-image match missed this one too.
CVE-2026-98959Patch Proof; Grype (kernel image)linux: 6.1.190-1 vulnerableSame. Grype reported it only through its NVD match on the kernel image file.
CVE-2026-99124Patch Proofvulnerable; note “[bookworm] - linux <ignored> (Architecture not supported in LTS)”An s390-only bug that Debian ignores in LTS. Patch Proof still lists it as “No distro fix yet”.
CVE-2019-15213Patch Proofvulnerable; “<postponed> (Revisit when correctly fixed upstream)”Open in every Debian release since 2019. Only Patch Proof reports it.
CVE-2026-31431Grype (kernel image)linux: fixed in 6.1.170-1 (DSA-6243-1)“Copy Fail”. The running 6.1.190-1 is fixed, but the old 6.1.128-1 kernel still installed is not. Grype flagged both kernel images; Patch Proof checks only the running kernel and said nothing.
CVE-2026-98464Grype (kernel image)bookworm-security 6.1.190-1: fixedFixed in exactly the running build. Grype read the version as 6.1.0-54 and compared it with upstream ranges.
CVE-2025-21692Grype (kernel image)linux: fixed in 6.1.128-1Both installed kernels are fixed. Grype flagged both.
CVE-2026-31608Grype (kernel image)bookworm <not-affected> (Vulnerable code not present)Never affected Debian 12.
CVE-2024-6387Grype (apt cache)openssh: fixed in 1:9.2p1-2+deb12u3 (DSA-5724-1); 1:9.2p1-2+deb12u10 installed“regreSSHion”, flagged from the image's original openssh-server_1:9.2p1-2 .deb. That file had sat in /var/cache/apt/archives since the provider built the image in March 2025.
CVE-2026-5773Grype (apt cache)curl: fixed in 7.88.1-10+deb12u15, the installed buildFlagged from the cached curl_7.88.1-10 .deb.
CVE-2023-45853Trivy, OSV-Scanner, Patch Proofzlib vulnerable, but “<ignored> (contrib/minizip not built and src:zlib not producing binary packages)”Debian says no zlib binary package is affected. Grype left it out; the other three, including Patch Proof, reported it.
CVE-2025-6020Trivy, OSV-Scanner, Patch Proofpam: vulnerableGrype didn't report it. Debian's record changed at 10:00 UTC that day (per OSV.dev's copy), after Grype's database was built at 06:19 UTC.
CVE-2025-8941OSV-Scanner, Patch Proofpam: undeterminedDebian hasn't triaged it. Trivy and Grype left it out; OSV-Scanner and Patch Proof listed it.
CVE-2026-106589OSV-Scanner, Patch Proofopenssh <not-affected> (Doesn't affect OpenSSH on Linux)Wrong in both. OSV.dev's copy, which both tools use, still listed Debian 12 as affected.
CVE-2026-19582OSV-Scanner, Patch ProofThe tracker page lists no package for itOSV.dev's copy still listed binutils in Debian 12 as affected.
CVE-2026-128886Trivy, OSV-Scanner, Patch Proofdbus: vulnerablePublished 13 Nov at 22:17 UTC, too late for Grype's database.
CVE-2026-124737OSV-Scanner, Patch Proofbusybox: vulnerableUpdated 14 Nov at 14:00 UTC, after Trivy's database was built (13:05 UTC). Grype didn't have it either.
CVE-2026-126820OSV-Scannercpio: vulnerablePublished 14 Nov at 14:47 UTC. Only OSV-Scanner, which queries OSV.dev live, had it. Patch Proof's 14:03 UTC snapshot didn't.

AlmaLinux 9.8 (web-02)

CVEReported byWhat the distribution’s data saysWhy the tools differ
CVE-2026-31431Trivy, GrypeAlmaLinux: fixed in kernel 5.14.0-687.5.3.el9_8 (ALSA-2026:19225); Red Hat: RHSA-2026:19225The running kernel is fixed. The old kernel-core 5.14.0-570.12.1.el9_6, still installed, is not, and nor is the 2022 rescue kernel. Patch Proof didn't mention either.
CVE-2025-38352Trivy, GrypeAlmaLinux: fixed in 5.14.0-570.42.2.el9_6 (ALSA-2025:15661)Same: affects only the old kernel package.
CVE-2025-21756Trivy, GrypeAlmaLinux: fixed in 5.14.0-570.17.1.el9_6 (ALSA-2025:7903)Same.
CVE-2025-9232GrypeRed Hat, RHEL 9: openssl “Fix deferred”There's no AlmaLinux advisory, because AlmaLinux publishes released fixes only. Trivy and Patch Proof can't show it.
CVE-2022-1616GrypeRed Hat: vim “Fix deferred”Same.
CVE-2023-51767GrypeRed Hat: openssh “Will not fix”Same; Grype labels it wont-fix.
CVE-2020-13844GrypeRed Hat: kernel “Will not fix”Same, on the running kernel.
CVE-2026-29111GrypeRed Hat: NetworkManager “Will not fix”; fixed in systemd (RHSA-2026:19213). AlmaLinux: systemd fix ALSA-2026:19213, installedOne CVE, two packages. Grype flags NetworkManager, as Red Hat does. Patch Proof counts the systemd fix as done.
CVE-2022-21123GrypeRed Hat: microcode_ctl “Affected”; kernel fixed in 5.14.0-162.6.1.el9_1 (RHSA-2022:8267)Same pattern: the fix shipped in the kernel.
CVE-2025-35973GrypeRed Hat: kernel “Affected”; fixed in microcode_ctl (RHSA-2026:68011), the build installedPatch Proof shows it fixed through microcode_ctl (ALSA-2026:68011).
CVE-2026-107364Grype (fix available)Red Hat: fixed in krb5-1.21.1-10.el9_9.2 (RHEL 9.9) and -10.el9_8.3 (RHEL 9.8 EUS)AlmaLinux had nothing newer than the installed 1.21.1-10.el9_8.1 (OSV.dev query, 19:23 UTC). A fix exists, but this server can't install it yet.
CVE-2026-35189Grype (fix available)Red Hat: fixed in openssl-1:3.5.9-1.el9_9Same: nothing newer than 1:3.5.8-2.el9_8 for AlmaLinux 9.
CVE-2026-23066Grype (fix available)Red Hat: fixed in kernel-5.14.0-741.14.1.el9_9Same, on the running kernel.
CVE-2026-6245Grype (fix available)Red Hat: fixed in sssd-2.9.9-3.el9Same.
CVE-2021-3640Grype (rescue kernel image)Red Hat and AlmaLinux: fixed in kernel 5.14.0-162.6.1.el9_1 (ALSA-2022:8267)The rescue kernel is that exact build, so it's fixed too. NVD's upstream range for “Linux 5.14.0” doesn't know that.
CVE-2022-0168Grype (rescue kernel image)fixed in 5.14.0-162.6.1.el9_1 (ALSA-2022:8267)Same.

Where Patch Proof fell short

Patch Proof is our tool, so here's the evidence against it, as plainly as the rest.

  • On AlmaLinux it can't see unfixed CVEs. It reported 0, while Red Hat lists 7,198 CVEs in these packages as affected with no fix, or won't fix: 512 in user space and 6,686 in the kernel. AlmaLinux's data has no such entries, which its README admits. If you need to know about known-but-unfixed issues on AlmaLinux, Grype was the only one of the four that showed them.
  • It ignores kernels that aren't running. All three of these can still be booted from GRUB, and Patch Proof said nothing about any of them:
    • Debian's old 6.1.128-1 kernel. It is affected by 2,444 CVEs that Debian fixed only in later builds, including “Copy Fail”.
    • AlmaLinux's old 570.12.1 kernel, where Trivy found 795 CVEs.
    • AlmaLinux's 2022 rescue kernel, affected by at least 680 CVEs that AlmaLinux fixed later.
  • Stale or noisy entries on Debian. Patch Proof's data comes from OSV.dev. About five hours after its snapshot, Debian's tracker no longer listed 12 of the CVEs it reported as affecting bookworm. 10 have no package entry; 2 are marked not affected, such as CVE-2026-106589 in OpenSSH. Patch Proof also reported CVE-2023-45853, which Debian ignores because no zlib binary is affected. Its “No distro fix yet” count doesn't separate Debian's <ignored> and <postponed> entries from plain open ones. It does show the “unimportant” urgency next to each CVE.
  • Its data was older. Its index was built at 14:10 UTC. OSV-Scanner, querying live at 18:44 UTC, found 4 Debian CVEs published or updated after that.
  • The problem it's built around barely showed up here. Patch Proof exists to stop backported fixes being reported as open. On both servers, Trivy, OSV-Scanner and Grype's package matcher compared installed packages with the fixed builds in the distribution's data (Red Hat's, for Grype on AlmaLinux). None of them reported a CVE as open because of a backported fix. The version-string matches the study found were Grype's kernel-image matches; the old .deb findings came from files that aren't installed packages.

Where it did help: it was the only tool that reported the running Debian kernel's open CVEs, and the only one that told the running kernel apart from the installed ones.

Which scanner is best for which job?

From each project's own documentation, plus what this test showed.

ToolBuilt forOn a Linux host, in this testWhat it downloaded
TrivyContainer images, filesystems, remote git repositories, VM images and Kubernetes. Finds vulnerabilities, IaC misconfigurations, secrets and licenses, and generates SBOMs. Skips third-party OS packages.The fewest CVEs on Debian (634), all consistent with Debian's data. Blind to Debian 12's kernel; flags old installed kernels on AlmaLinux. No unfixed CVEs on AlmaLinux.134 MiB vulnerability DB, plus a 953 MiB Java DB on Debian because .jar files exist under /usr/share/java
GrypeContainer images, filesystems and SBOMs, with OS and language packages. EPSS, KEV and risk scores, and OpenVEX filtering. Pairs with Syft for SBOMs.The widest net: unfixed Red Hat statuses, rescue and old kernels, cached .deb files. Also the most findings to sort through: its version-string kernel matches added 6,112 CVEs on Debian, of which 708 are open and 2,444 still affect the older installed kernel.Database: 3.3 GB on disk. Needed over 1.4 GB RAM on AlmaLinux.
OSV-ScannerLockfiles and manifests in source trees (11+ language ecosystems), container images (Alpine, Debian, Ubuntu packages), guided remediation and license checks. A front end for the OSV.dev database.Debian: read the dpkg database and matched Trivy closely. AlmaLinux: found nothing to scan. No kernel CVEs.Nothing; queries the OSV.dev API (an offline database is optional)
Patch ProofDistribution packages on one Debian, Ubuntu, AlmaLinux or Rocky Linux host. Classifies each CVE: fixed by backport, open, no fix yet, or patched on disk but not running. No containers or language packages.The running Debian kernel's open CVEs, which the others missed. Blind to non-running kernels and to unfixed CVEs on AlmaLinux.41 MB index, on your machine; only a read-only shell script runs on the server

If you scan container images or infrastructure code, Trivy or Grype is the tool. For a single fleet of Linux servers, run one generic scanner and one distribution-aware check, then read the kernel lines carefully.

How we tested

  • Servers. Two VMs with 2 vCPUs and 1.7 to 1.9 GB of RAM each, built from a hosting provider's templates and fully patched earlier that day (16:44 UTC on Debian, 17:57 UTC on AlmaLinux): Debian 12.15 and AlmaLinux 9.8.
  • Tools. Each was installed from its GitHub release:
    • Trivy 0.76.2 as its .deb or .rpm
    • Grype 0.121.3 as its tarball
    • OSV-Scanner 2.7.1 as its binary
    Each file was checked against the project's published SHA-256 list, on a separate machine and again on each server. Patch Proof 1.0.0 was installed on a separate machine from its v1.0.0 tag, using the README's install command with pip in a virtual environment instead of pipx. Only its read-only collector ran on the servers. No curl | sh.
  • Data dates.
    • Trivy DB: updated 14 Nov 13:05 UTC.
    • Grype DB: v6.1.10, built 14 Nov 06:19 UTC.
    • OSV-Scanner: the OSV.dev API, live at scan time.
    • Patch Proof index: built 14 Nov 14:10 UTC from OSV.dev's Debian data of 14:03 UTC and AlmaLinux data of 13 Nov 16:03 UTC.
  • Modes. Defaults, apart from the flags below. Trivy and Grype skipped /proc, /sys, /dev, /run, their own files and other tests' directories.
    • trivy rootfs --scanners vuln /, the documented host mode
    • grype dir:/
    • osv-scanner scan source -r over /boot /etc /srv /usr /var, its documented directory mode; it has no documented host mode
    • hc-proof collect piped over SSH, then hc-proof report
  • Other tests on the same VMs. Other software tests were running on both servers at the same time. We hashed the package database before and after every scan and used only runs whose hash matched the baseline. We also checked that every tool saw the same installed packages. On AlmaLinux, another test's install overlapped the first Trivy and OSV-Scanner runs, so we discarded and repeated them.
  • Counting. Unique CVE IDs per tool and host. Advisory IDs (DEBIAN-CVE, ALSA, GHSA) were mapped to their CVE aliases.
  • Afterwards. We uninstalled every scanner and deleted its caches and databases. Nothing else was changed, and nothing was rebooted.

Limits of this test

  • Two servers, one day, one run per tool. Different packages, distributions or days will give different numbers.
  • Both servers were fully patched. A server with missing updates would also produce “fix available” findings. We didn't test that case.
  • Hosting-provider images, not official cloud images. The leftover old kernels, the 2022 rescue kernel and the populated apt cache are partly artifacts of those images.
  • Default settings only. Each tool has filters that change the counts, such as Trivy's --ignore-unfixed, Grype's --only-fixed and OSV-Scanner's “unimportant” filter.
  • Databases built hours apart. Some small differences are timing, not method.
  • The AlmaLinux reasons were spot-checked, not checked per CVE. We checked every Debian disagreement against the tracker. On AlmaLinux we checked the 16 sampled CVEs individually, and the 7,198 “Red Hat says affected” count comes from Grype's own Red Hat match.

How to reproduce it

On a test server, as root. Use your own paths and check each tool's current release first.

# 1. Download a pinned release and check it (Grype shown; Trivy and OSV-Scanner work the same way)
curl -fsSLO https://github.com/anchore/grype/releases/download/v0.121.3/grype_0.121.3_linux_amd64.tar.gz
curl -fsSLO https://github.com/anchore/grype/releases/download/v0.121.3/grype_0.121.3_checksums.txt
grep ' grype_0.121.3_linux_amd64.tar.gz$' grype_0.121.3_checksums.txt | sha256sum -c

# 2. Scan
trivy rootfs --scanners vuln --format json --output trivy.json /
grype dir:/ -o json --exclude './proc/**' --exclude './sys/**' --exclude './dev/**' --exclude './run/**' > grype.json
osv-scanner scan source -r --format json /boot /etc /srv /usr /var > osv.json
# Patch Proof runs on your own machine and reads the server over SSH:
hc-proof collect --script | ssh root@server sh -s | hc-proof report --input - --out report.html --json patch-proof.json

# 3. Count unique CVE IDs
jq -r '.Results[] | select(.Class=="os-pkgs") | .Vulnerabilities[]?.VulnerabilityID' trivy.json | grep '^CVE-' | sort -u | wc -l
jq -r '.matches[] | .vulnerability.id' grype.json | grep '^CVE-' | sort -u | wc -l
jq -r '.results[]?.packages[].vulnerabilities[].upstream[]?' osv.json | sort -u | wc -l
jq -r '.findings[] | select(.class=="open" or .class=="notrunning" or .class=="nofix") | .id' patch-proof.json | sort -u | wc -l

Grype's JSON tells you where each match came from:

  • .artifact.locations[0].path separates the package database from boot/vmlinuz-* and var/cache/apt/archives/*.deb.
  • .vulnerability.namespace shows whose data it used.

To check one CVE by hand:

  • Debian: open https://security-tracker.debian.org/tracker/CVE-… and compare the bookworm line with dpkg-query -W.
  • AlmaLinux: compare the fixed build in AlmaLinux's errata with rpm -q.

Our guide to checking for security updates covers the apt side.

How we checked this: Two test VMs built from a hosting provider's images and fully patched that day: Debian 12.15 and AlmaLinux 9.8. Scanned on 14 Nov 2026, 18:41 to 19:20 UTC, with Trivy 0.76.2 (trivy rootfs), Grype 0.121.3 (grype dir:/), OSV-Scanner 2.7.1 (scan source) and HostCrust Patch Proof 1.0.0. Each was installed from its official GitHub release with its checksum verified, and removed afterwards. Every CVE in the two sample tables was looked up the same day in Debian's security tracker or Red Hat's security data, and in AlmaLinux's data through OSV.dev. Anything we didn't run ourselves comes from the official documentation listed under Sources.

Check which CVEs are really open on your server

HostCrust Patch Proof compares every installed package with your distribution's own security data. It counts fixes your distro backported and checks the kernel that is actually running, not just the one on disk. Read-only, no root, free.

Run the free check About Patch Proof

Run this check on every server, every day

HostCrust connects your servers with a small outbound-only agent and runs Patch Proof on each one daily. Free for up to 3 servers.

See how HostCrust works

FAQ

Is Grype or Trivy better for scanning a Linux server?

For installed user-space packages on our Debian 12 server they matched almost exactly: all 630 CVEs Grype reported were among Trivy's 634. They split on two things:

  • The kernel. Trivy reported no kernel CVEs on Debian 12. Grype added 6,112 by matching the kernel image files in /boot against NVD version ranges.
  • The data source on AlmaLinux. Trivy uses AlmaLinux's errata, which list released fixes only. It reported 795 CVEs, all on an old kernel package that wasn't running. Grype used Red Hat's data and reported 12,598, including 7,198 that Red Hat lists as affected with no fix, or won't fix.

Choose by what you need to see: Trivy reported far fewer CVEs, and Grype is the only one of the two that showed unfixed CVEs on AlmaLinux.

Why does Grype show thousands of kernel CVEs on a patched Debian server?

Grype catalogued the kernel images in /boot as files and matched them against NVD's upstream version ranges. It read their versions as 6.1.0-54-amd64 and 6.1.0-31-amd64, which is Debian's kernel ABI name, not the 6.1.190 upstream release that's actually running. Of the 6,112 CVEs it matched:

  • 5,262 are fixed, per Debian's security tracker, in a build no newer than the running 6.1.190-1.
  • 136 never affected Debian 12.
  • 708 are really open in Debian.
  • 6 aren't in Debian's tracker under linux.

2,444 of the fixed ones do still affect the older 6.1.128-1 kernel left installed on the server.

Can OSV-Scanner scan the packages installed on a Linux server?

Partly, and not officially. OSV-Scanner documents OS-package scanning for container images (Alpine, Debian, Ubuntu). It has no documented host mode. When we ran its documented directory mode, osv-scanner scan source -r, over the host:

  • Debian 12: it read /var/lib/dpkg/status and reported 653 CVEs.
  • AlmaLinux 9.8: it stopped with “No package sources found”, because it doesn't read the RPM database in that mode.
Why did Trivy report no kernel CVEs on Debian 12?

Trivy looks up Debian advisories by the installed package's source name. Debian 12's linux-image-* packages are built from the source package linux-signed-amd64, and Debian's tracker files kernel CVEs under linux. So the lookup finds nothing. OSV-Scanner and Grype's package matcher found no kernel CVEs either.

Patch Proof was the only tool that reported the running 6.1.190-1 kernel's CVEs with no Debian fix: 1,994 of them. When we checked Debian's tracker about five hours later, 1,984 were still open.

Why does a fully patched Linux server still show hundreds of CVEs?

Because the distribution hasn't fixed them. All four tools agreed on 630 CVEs in Debian's installed user-space packages, and none had a Debian fix. Debian's tracker gives them these statuses:

  • 228 have urgency “unimportant”.
  • 145 are <postponed>.
  • 76 are <no-dsa>: too minor for a security advisory.
  • 10 are <ignored>.
  • 171 are open with no decision recorded yet.

Installing updates can't remove these. To act on them, read the tracker's notes, then mitigate (for example, remove the package or switch off the feature) or accept the risk.

What is a good Trivy alternative for Linux servers?

It depends on the job:

  • Grype covers the same ground for images, filesystems and SBOMs.
  • OSV-Scanner is strongest on lockfiles and container images.
  • HostCrust Patch Proof covers only the distribution packages on a Debian, Ubuntu, AlmaLinux or Rocky Linux host. It uses the distribution's own data and checks the running kernel.

On our AlmaLinux server Patch Proof reported nothing unfixed, because AlmaLinux's data has no unfixed entries. Grype did report them.

Sources

  1. Trivy v0.76.2 README: targets and scanners
  2. Trivy docs: rootfs target (host machine scanning)
  3. Trivy docs: vulnerability data sources and vendor-advisory selection
  4. Trivy docs: AlmaLinux coverage (unfixed vulnerabilities not supported)
  5. Trivy source: the Debian detector looks advisories up by source package name
  6. Grype v0.121.3 README: scan targets, EPSS/KEV, OpenVEX
  7. Anchore docs: supported operating systems and vulnerability data sources
  8. Anchore docs: Syft scan targets (directory scans of a whole system root)
  9. OSV-Scanner v2.7.1 README
  10. OSV-Scanner docs: supported artifacts and lockfiles
  11. OSV-Scanner docs: scanning a source directory
  12. OSV-Scanner docs: container scanning (Debian 'Unimportant' entries filtered in table output)
  13. HostCrust Patch Proof v1.0.0 README: classes, kernel handling, known data gaps
  14. Debian security tracker (per-CVE status by release)
  15. Debian security tracker data in JSON
  16. Red Hat CVE database (per-CVE package states and fixed builds)
  17. AlmaLinux errata