Jenkins CVE-2026-70427

Exploitation requires the attacker to control a Jenkins agent process - either by directly compromising an agent host at the OS level, by injecting malicious code into a build running on an agent, or by having Agent/Connect permission in Jenkins and operating a rogue agent. The attack does not require any direct HTTP access to the Jenkins controller web UI or administrator credentials. Exploitation is bounded by the file system permissions of the OS user account running the Jenkins controller process; writes to paths outside JENKINS_HOME or outside directories writable by that user will fail. Controller-only Jenkins deployments with no connected agents are not affected. There is no reported requirement for a specific non-default Jenkins configuration beyond the presence of at least one connected agent.

No CVSS vector or EPSS score was provided in the input data, so risk signals must be inferred from the vulnerability description. The attack requires control over a Jenkins agent process - a meaningful but non-trivial prerequisite that nonetheless represents a realistic threat model in multi-team or externally-integrated Jenkins deployments where agents execute untrusted build code. The impact ceiling is severe: a write-to-arbitrary-path primitive on the controller that the advisory explicitly maps to code execution via Groovy init scripts or plugin deployment. The constraint is OS-level file permissions of the Jenkins user; installations where Jenkins runs as a low-privilege OS user limit writable paths, but JENKINS_HOME itself is always writable. No KEV listing, no confirmed POC, and no EPSS data are available to calibrate real-world exploitation probability. The risk is materially elevated for organizations with shared Jenkins infrastructure, third-party agent integrations, or pipelines processing untrusted artifacts.

An attacker who has compromised a Jenkins build agent - or who has code running on an agent through a malicious dependency or build script - crafts a .tar.gz archive containing a symbolic link entry with an effectively empty name. When the archive is transferred to and extracted by the Jenkins controller, the symlink causes a file to be written outside the intended extraction directory, placing a malicious Groovy script in JENKINS_HOME/init.groovy.d/. On the next Jenkins restart or reload, the controller executes the planted script, granting the attacker arbitrary code execution with the privileges of the Jenkins process. No public proof-of-concept has been identified at time of analysis.