CISA added CVE-2026-97229 to KEV on October 29: the CI/CD regex compiler flaw allows unauthenticated code execution in affected GitLab CE and EE versions. Install the latest supported branch patch and review vendor detection guidance.
Who should check: Teams operating GitLab CE or EE 19.2, 19.3 or 19.4 on their own infrastructure.
Recommended actionFix available
Install the latest supported branch patch, verify the running version and investigate the CI/CD indicators in the vendor advisory. Do not downgrade to an older minimum fix.
An integer overflow in GitLab’s CI/CD regular expression compiler can allow a remote attacker to execute code without authentication. GitLab rates CVE-2026-97229 at CVSS 3.1 10.0. CISA added this exact vulnerability to KEV on October 29; the entry does not establish when attacks began.
GitLab published the original fixes on October 12: 19.2.8, 19.3.5 and 19.4.2. These are historical minimum fixes, not a recommendation to install an older patch today. Use the latest patch for a supported branch, including corrections for subsequent vulnerabilities. GitLab says GitLab.com and GitLab Dedicated were patched.
The vendor advisory provides six detection scenarios covering native regex worker crashes, hostile GraphQL CI lint expressions, oversized non-ASCII paths, guard rejections, Workhorse worker crashes and scripted client error bursts. Use their documented conditions when investigating; a crash alone does not prove exploitation.
This is separate from CVE-2026-110264. Their similar descriptions do not justify transferring KEV evidence between them. Keep both vulnerabilities in the patch review and verify the version actually serving traffic.
Vendor / publisher
GitLab
Vulnerability identifiers
CVE-2026-97229
Access required
Application or configuration dependent
Vendor severity
10.0 · CVSS 3.1
Disclosed
Oct 12, 2026
Latest evidence check
Oct 30, 2026 UTC
Which systems and versions are affected?
Product
Affected
Fixed release / action
GitLab CE / EE 19.2
19.2 before 19.2.8
19.2.8 — original minimum fix; use the latest supported branch patch, including later security fixes.
GitLab CE / EE 19.3
19.3 before 19.3.5
19.3.5 — original minimum fix; use the latest supported branch patch, including later security fixes.
GitLab CE / EE 19.4
19.4 before 19.4.2
19.4.2 — original minimum fix; use the latest supported branch patch, including later security fixes.
Use the guidance for your product and release branch. Confirm distribution backports with the vendor; the product name alone does not establish exposure.
CISA lists this exact CVE as exploited. Affected self-managed GitLab releases allow unauthenticated code execution through the CI/CD regular expression compiler.
What does this mean for AI environments?
Review repository tokens, CI/CD secrets and runner credentials accessible to a compromised instance. This is an exposure review, not evidence that an AI agent caused exploitation.
Added the October 29 KEV inclusion for CVE-2026-97229 and the original October 12 branch fixes. Kept this evidence separate from CVE-2026-110264 and directed operators to later supported patches.
Response actions
Plan your response
0 of 4 actions recorded as done
Use this checklist to decide whether the advisory affects your systems, record what your team does and share the result. Radar does not scan or verify your environment.
Changes save automatically in this browser. They are not synced with your team or sent to Oktsec.
Name the system and person or team handling the work. Update each action as you go; add a ticket reference or notes to explain what was checked.
1
Inventory deployed CE and EE versions
Identify affected 19.2, 19.3 and 19.4 instances. Verify the running version on each node rather than only the intended package or image.
Add evidence or notesUse references only. Do not store credentials or confidential data here.
2
Install the latest supported branch patch
The listed versions are the original October 12 fixes. Later updates fix additional vulnerabilities, including CVE-2026-110264. Follow supported upgrade paths and do not downgrade to these historical minimum versions.
Add evidence or notesUse references only. Do not store credentials or confidential data here.
3
Use the vendor detection guidance
Review CI lint and regular expression worker crashes, hostile regex requests, oversized non-ASCII paths and guard rejections using the six detection scenarios in the advisory. Preserve logs before maintenance; patch completion does not establish absence of compromise.
Add evidence or notesUse references only. Do not store credentials or confidential data here.
4
Review credentials reachable from the instance
If investigation identifies compromise, scope access to repositories, CI/CD secrets, runners and integrations before revoking or rotating affected credentials. Record findings and telemetry gaps.
Add evidence or notesUse references only. Do not store credentials or confidential data here.
3. Share the plan
Download a Markdown file with your decision, action status, notes and source links. Share it with your team or attach it to a ticket.
Apply mitigations in accordance with vendor instructions, ensuring compliance with CISA’s BOD 26-04 Prioritizing Security Updates Based on Risk (see URL in Notes) guidance and CISA’s “Forensics Triage Requirements” (see URL in Notes). Follow applicable BOD 26-04 guidance for cloud services or discontinue use of the product if mitigations are unavailable. Stakeholders are responsible for evaluating each asset's internet exposure and ensuring adherence to BOD 26-04 patching guidelines.
CISA’s federal deadlines and requirements are shown for context. They are not a universal deadline for private organizations or businesses in Latin America. The date added is the catalog inclusion date, not the first attack date.
Priority applies to the parent advisory, not separately to every product or CVE. Each row shows only the evidence attributed to that CVE. Missing attack evidence does not mean a vulnerability is safe to ignore. Your environment has not been scanned.
Review repository tokens, CI/CD secrets and runner credentials accessible to a compromised instance. This is an exposure review, not evidence that an AI agent caused exploitation.