The CVSS 6.3 score for CVE-2026-44517 undersells the real risk. This vulnerability in Buildah's TempDirForURL creates a path traversal that can be triggered not only through a hypothetical 'malicious server' but through the tool's intended, documented workflow: pulling Git repositories as build contexts. When your CI pipeline does exactly what Buildah advertises — clone a repo, build an image — you're crossing a trust boundary the tool never signals to you.

The vulnerability has two components. First, Git repository subdirectory handling can leak files outside the intended build context. Second, the stdinToDirectory function can follow symlinks during tar extraction, allowing path traversal when partially-extracted archives exist on the filesystem. These aren't separate bugs; they're symptoms of a single architectural weakness: Buildah treats the build context directory as a secure container without enforcing that boundary during extraction and context assembly.

Patch versions 1.43.2 and 1.44.0 both contain the fix, which is unusual — coordinated releases across active branches suggest either a coordinated downstream disclosure or internal discovery during a security audit. The simultaneous landing makes a surgical fix (path canonicalization checks) more likely than a broader architectural refactoring. This matters because it means other extraction paths in Buildah may have similar issues not yet identified.

Prioritize remediation if you use Buildah in automated pipelines with Git-based build contexts, regardless of the medium CVSS rating. The practical exploitation landscape is broader than the scoring suggests because the attack surface includes any compromised Git hosting platform, MITM on CI/CD pipelines fetching repos, or malicious internal 'upstream' repositories — scenarios the CVSS 'malicious server' vector doesn't capture.

Audit your Buildah and related Podman codebases for other context-fetching paths that assume the extracted directory is a secure boundary. Treat this as a supply chain vulnerability: if exploited, the poisoned image propagates downstream with signed legitimacy, into Kubernetes clusters and production systems running Podman.