Hello @osobo
Because of all the new CVEs thanks to LLMs I have started to use https://dependencytrack.org/.
I am still only tinkering with it and validating if we can use it in our Cloudron development and packaging workflows.
With that information, I have checked my dependency track dashboard for cloudron/python-base tag 3.12-20261023 and cloudron/base tag 6.0.1.
@osobo said:
Docker Scout reports 10 critical and 74 high findings
Interesting!
Dependency track shows the following for the base image:
[image: deptrack.cloudron.dev_projects_f21919f2-20d4-4a9f-96c5-6e379ee18096_findings.jpeg]
So your report from Docker Scout is 10 critical, but dependency track found 28 critical, 134 high, 119, medium, 41 low and 71 unassigned.
Dependency track also has a view for Exploit Predictions with a KEV value if a CVE is actively exploited in the wild.
From that data I have from decency track my findings are the following:
None of the CVEs in cloudron/base:6.0.1 or cloudron/python-base:3.12-20261023 are on the CISA KEV catalog or the ENISA EU KEV list. Nothing is known to be exploited in the wild.
The EPSS scores are low. The highest is 0.065 (6.5% chance of exploitation activity in the next 30 days). 461 of 489 findings score below 0.01 (1%).
Top 5 by EPSS:
EPSS
ID
Package
0.065
UBUNTU-CVE-2026-123633
nginx 1.28.3-2ubuntu1.11
0.064
UBUNTU-CVE-2026-123007
nginx 1.28.3-2ubuntu1.11
0.054
CVE-2007-0770
graphicsmagick 1.3.46
0.040
CVE-2026-45447
openssl 3.5.5-1ubuntu3.7
0.023
CVE-2025-61726
Go stdlib 1.24.6
The GraphicsMagick entry is a 2007 CVE matched against a 2025 release, so most likely a false match.
cloudron/python-base:3.12-20261023 only adds 8 packages on top of cloudron/base:6.0.1 (python3.12 3.12.15 and pip 26.2.2), and Dependency-Track has no findings for them. The python3.12 packages are not from the Ubuntu archive (version suffix +resolute1), so the Ubuntu advisories may not cover them.
Why the numbers differ from Docker Scout?
Dependency-Track only sees the final filesystem of the image. Binaries deleted in a later layer, like the gosu, yq and pebble you mentioned, do not show up at all. Scout reports them from the lower layers. The vulnerability sources also differ, so the counts are not directly comparable.
So yes, there are CVEs, but nothing known exploited and nothing with a high exploit prediction. They get fixed with the regular base image updates.
The two nginx findings at the top are both HTTP/2 bugs (CVE-2026-123633, CVE-2026-123007).
On Cloudron the host nginx terminates TLS and HTTP/2 and proxies to the app container over HTTP/1.1, so the nginx inside the container never receives HTTP/2 traffic. For these two, the host nginx is the one that matters, and it is updated through the host OS, not the base image.
Why a CVE in an image does not automatically mean a Cloudron app is affected
A scanner lists every package with a known CVE.
Whether that CVE can actually be used against a Cloudron app depends on how the app runs, in this case for the NGINX CVE:
Reverse proxy: the host nginx handles TLS and HTTP/2 and passes requests to the app container over plain HTTP/1.1. Bugs in the TLS or HTTP/2 code of a package inside the container (like the two nginx CVEs above) cannot be reached that way
Container hardening: apps run with a read-only root filesystem, a restrictive AppArmor profile and their own Docker network
Host updates: Cloudron enables unattended-upgrades on the host, so Ubuntu security fixes for the host (including the host nginx) are installed automatically
Each of these layers narrows which vulnerabilities are reachable at all, which is why scanner counts alone overstate the real risk. It also makes checking a single CVE more work, because the answer depends on the package, how it is used, and which layer sits in front of it.
These layers do not protect against bugs in an app's own code, since those arrive inside normal requests. Those are fixed by updating the app package, which is why keeping apps updated (automatic updates are on by default) matters most.
What we are doing
I am continuing the evaluation of Dependency-Track for Cloudron, so we can see which images are affected by new CVEs, which of those are known to be exploited (KEV) and which have a high exploit prediction (EPSS). The goal is to fix the ones that matter first, instead of sorting through hundreds of scanner findings by hand.
It is still early, but it already helped to answer this thread with data instead of guesses.
If you find something that looks exploitable on Cloudron, please report it privately to security@cloudron.io.