CVE-2025-53618 is a high severity (CVSS 7.4) out-of-bounds read in Grassroots DICOM, also called GDCM, version 3.024. A specially crafted DICOM file can trigger it in the JPEGBITSCodec::InternalCode function and lead to an information leak. Check the Talos report and the project for a fixed release.
What's affected
The record describes an out-of-bounds read in the JPEGBITSCodec::InternalCode functionality of Grassroot DICOM 3.024. It says a specially crafted DICOM file can lead to an information leak and that an attacker can provide a malicious file to trigger the flaw. The weaknesses are CWE-119 and CWE-125.
The record adds that the function grayscale_convert is called based on a value in the malicious file that specifies how the image pixel data should be interpreted. In other words, a field in the file steers the decoder into a conversion path that reads past the end of a buffer.
The reference is Talos report TALOS-2025-2210. I could not load that page while preparing this article, so a fixed version is not confirmed here. Check the Talos report and the GDCM project releases for the fix.
How to check if you're exposed
Find where GDCM is used. It is a library for reading DICOM medical images, so it appears inside viewers, research tools, PACS gateways, image pipelines and Python or Java bindings. Search your dependency manifests and container images:
pip list 2>/dev/null | grep -i gdcm
find / -name "libgdcm*" 2>/dev/null
A version of 3.024 is named in the record. Also look for tools in your environment that bundle GDCM without saying so, such as imaging research software.
Then map where DICOM files come from. A pipeline that decodes files from external partners, uploads or removable media is more exposed than one that only reads files from your own scanners.
What to do now
- Find every application and image that includes GDCM and note the version.
- Read Talos report TALOS-2025-2210 and the GDCM project releases for a fixed version, and update when one is available.
- Do not open DICOM files from unknown sources in affected tools.
- Run DICOM decoding in a sandboxed or low privilege process, away from patient data stores where possible.
- Rebuild container images that bundle the library after you update, and scan again to confirm.
Why it matters
An out-of-bounds read means the decoder reads memory beyond a buffer, and in an image decoder that happens when a header field tells the code to expect a different layout than the data really has. The result can be a crash or a disclosure of memory contents. The record scores the impact as high on all three measures, with a local attack vector, so the attacker needs a file to be opened by a user or a process on the host.
Medical imaging software handles sensitive data and often processes files from outside parties. The CVE is not on the CISA Known Exploited Vulnerabilities list, and the record gives no sign of attacks. Dependency scanning is the practical way to find a library that is buried inside several applications.
Quick check
Do I need a login to be affected?
The record uses a local attack vector with no privileges. The risk comes from a malicious file being opened by your software.
Sources
Last reviewed: 18 October 2026.
