Build1 publisher3 min readPublished
Seven sandbox bypasses in the September Jenkins advisory all terminate in the same plugin
All seven end in arbitrary code execution on the controller and all are fixed in one Script Security release, so the work is finding out which version of it each of your controllers actually runs.
The Engineer · Build desk

What happened
- The Jenkins advisory dated September 16, 2026 covers 20 plugin vulnerabilities, and seven of them break the Script Security sandbox to give code execution on the controller.
- The Gradle and Bitbucket plugins carry SSRF issues that can capture stored credentials, and the guide pairs their upgrades with rotating any credential those jobs can reach.
- The Jenkins project reported no known active exploitation and no public proof-of-concept as of publication.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- constraint Remediation runs at plugin-version granularity, so a core upgrade schedule buys nothing here and the patch window has to be driven off an inventory most teams keep in a UI that only records the last click.
- decision Each operator now has to decide between patch-and-verify and assume-compromise, and the deciding input is how many people hold Pipeline definition rights.
- exposure For anyone running global credentials, a sandbox bypass means every global credential is disclosed, and the blast radius is the agents the controller can reach.
- cost For installations that cannot rule out exploitation, remediation means credential rotation plus re-verification of every artifact built in the window, and the teams downstream of those builds do that work.
A sandbox bypass needs somewhere to run. The attacker has to be able to define a Pipeline and start it, and according to a dev.to hardening guide that precondition holds for every one of the seven bypasses in this advisory [13]. Where Pipeline definition and job configuration rights sit with a small audited group, the class of attack has no entry point [13]. The guide calls that restriction "the single highest-leverage configuration change available" [13].
All seven bypasses terminate in Script Security, fixed in 1422.v06869826dd9b_, with the affected range running up to 1415.v9af9b3ac253d [4]. So one plugin upgrade closes 35 percent of the advisory: seven of the 20 entries, leaving 13 spread across the other components [1]. Two of those, CVE-2026-92134 and CVE-2026-92135, list fixed-version strings that differ from the rest, so a uniform bump will miss them [8]. The guide does not reproduce the version table; it tells you to read it [8].
Every flaw in the advisory lives in a plugin [3]. The named components are Script Security, Robot Framework, Gradle, Bitbucket, Warnings, Coverage and Dependency-Check, which the guide describes as common enough that most mature installations run at least one [18].
The script console will enumerate installed plugins, and the update center shows what updates exist [9]. For larger installations the guide prefers configuration management or a plugin inventory export, on the grounds that what is deployed and what someone last clicked in the UI are two different things [10]. The same problem appears on the way out: plugin updates applied through automation fail quietly more often than teams expect, and from the dashboard a failed update looks identical to a successful one [11].
Robot Framework is a separate job, fixed in 6.3.0, affected up to 6.2.2, with a path traversal flaw that can escalate to remote code execution [5]. Gradle and Bitbucket carry SSRF issues that can capture stored credentials, so the guide pairs the upgrade with rotating any credential those jobs can reach [6]. Warnings, Coverage and Dependency-Check are stored XSS, lower severity, same maintenance window [7]. Global credentials handed to every job turn one controller-level compromise into disclosure of all of them, and because the controller can reach every agent, agent trust boundaries decide how far that compromise travels [14][15].
The Jenkins project reports no known active exploitation and no public proof-of-concept as of publication [12]. The guide takes that as the vendor's assessment at disclosure time, made without reference to any particular environment [12].
Whether patching is enough turns on the length of the exposure window and how widely Pipeline definition rights are granted [16]. Broad rights, or affected plugins installed for a long period without version tracking, argue for rotating every credential reachable from a controller-level job, comparing job definitions against version control, reviewing build history for jobs that ran outside normal patterns, and holding artifacts built in the window for independent verification before distribution [16][17]. For the next advisory, the guide names three controller events to log: job definition changes, plugin installations, and script console usage [19].
What to watch
- A public proof-of-concept for any of the seven bypasses would move triage from patch-and-verify to assume-compromise.
- Whether the odd entries, CVE-2026-92134 and CVE-2026-92135, end up with fixed versions consistent with the rest of the table.
- Any revision by the Jenkins project to its no-known-exploitation statement.