On this page (7 sections)
A critical flaw has been published in imgproxy. According to the GitHub Advisory Database entry, imgproxy from 4.0.0 before 4.0.16 contains an integer overflow in its ICO loader that can lead to a heap buffer overflow when a crafted ICO image is processed.
If you run imgproxy, or manage a site or platform that does, check your version now. The advisory was published on November 8, 2026, and so far only one independent publisher covers it, so details may still change.
What happened
The advisory describes an integer overflow in the ICO image loader. A remote attacker can supply an ICO file with a crafted data_size value. That value makes the allocation size wrap around, so the memory allocated is smaller than the data later written into it. The result is heap memory corruption, which the advisory says can crash the process or achieve code execution.
The advisory names vips/icoload.c as the affected code. The research does not explain that component further or say whether any other software is affected, so do not assume either way.
Technical details
An integer overflow happens when arithmetic on a size produces a number too large for its type, and the value silently wraps to something small. When that wrapped number is used to allocate a buffer, the program believes it has enough room, but the real data is larger. Writing it spills past the end of the buffer into neighboring heap memory. This is the weakness class CWE-122.
The CVSS v4 metrics in the GitHub entry describe the exposure:
- Attack vector: network, so the attacker does not need local access.
- Attack complexity: low, with no special attack requirements.
- Privileges required: none, and no user interaction is needed.
- Impact on the vulnerable system: high for confidentiality, integrity and availability.
- No impact is scored for subsequent systems.
Who is affected
Only deployments running imgproxy 4.0.0 through 4.0.15 are in scope according to the advisory text. The CVSS metrics say the attack is network-based with no privileges needed, so any instance that can be made to process an attacker-supplied ICO image should be treated as exposed. The research does not say which configurations make that possible.
As general guidance, not something the advisory states, give priority to instances that are reachable from the internet or that process images from sources you do not control. Instances that only handle images you fully control are likely harder to reach, but they are still within the affected version range and should be upgraded.
How to check your exposure
First find every place imgproxy runs: containers, Kubernetes deployments, VMs and any staging or preview environments. Then confirm the version. If you run it in Docker, the following command lists local images that mention imgproxy, with their tags, so you can compare against 4.0.16. It is read-only.
docker images | grep -i imgproxyIf you deploy through Kubernetes, this prints each pod's container images across all namespaces so you can spot imgproxy tags below 4.0.16. It only reads cluster state.
kubectl get pods --all-namespaces -o jsonpath='{range .items[*]}{.metadata.namespace}{"\t"}{.metadata.name}{"\t"}{.spec.containers[*].image}{"\n"}{end}' | grep -i imgproxyA tag such as latest does not tell you the version. In that case check your deployment records or image digest, and when in doubt, redeploy a build you have confirmed is 4.0.16 or newer.
What to do now
The first two items follow from the advisory's affected range. The rest are general hardening suggestions, not guidance from the advisory.
- Inventory every imgproxy instance, including staging and internal ones.
- Upgrade anything running 4.0.0 to 4.0.15 to 4.0.16 or later. The fixed version is implied by the affected range, so confirm it against the referenced GitHub security advisory GHSA-hff9-5pqw-6w9f or the project's release notes.
- Pin the image tag to a specific version instead of latest so you can verify what is running.
- Check the referenced advisories for any mitigation the maintainers describe. Their contents were not available to us, so we cannot say whether any exist.
- As general hardening, limit who can reach imgproxy and, where your setup allows, restrict which sources it fetches images from. The research does not say whether this prevents exploitation.
- As general hardening, run imgproxy as an unprivileged user with minimal network and filesystem access.
- Review crash logs and restarts from recent weeks for unexplained process failures.
Detection and monitoring
The research does not describe specific indicators of exploitation, so there is no reliable signature to hunt for. Practical signals are indirect: repeated crashes or restarts of the imgproxy process, out-of-memory or segmentation fault messages, and requests for ICO files from unfamiliar sources. On Kubernetes, the following lists pods with their restart counts, so you can spot imgproxy containers that keep crashing. It is read-only.
kubectl get pods --all-namespaces | grep -i imgproxyIf you find a crash you cannot explain, or any sign that the process executed something unexpected, treat the host as potentially compromised. Rotate any secrets and keys the container could read, such as signing keys and storage credentials, and follow our guide on what to do if your website is hacked.
For a broader view of how your public site looks from the outside, a free malware scan and the guide to checking if a website is hacked are useful follow-ups, though an outside scan cannot see inside a private image service.
What we still do not know
- Whether the flaw is being exploited. The research reports no exploitation and shows no CISA KEV entry, but that is not confirmation either way.
- Whether a public proof of concept exists. None is mentioned.
- Exact patched version details beyond the affected range. The GitHub entry lists patched versions as unknown, though the range implies 4.0.16.
- Which deployment configurations are practically reachable, and whether any other software is affected.
Treat this as a developing story until the referenced advisories are reviewed. Even without known exploitation, a network-reachable memory corruption bug scored 9.3 that needs no privileges is worth upgrading for promptly.
Sources
This article is based on the public sources below. Details can change as vendors and researchers publish more; check the original advisories before acting on version numbers.
