# CVE-2026-96893: GitLab EE prompt injection in AI agent instructions

Source: https://junglewise.ai/threats/cve-2026-96893-gitlab-has-remediated-an-issue-in-gitlab-ee-affecting-all  
Publisher: Junglewise Threat Intelligence, https://junglewise.ai/threats  
Published: 2026-10-13. Updated: 2026-10-16  
License: CC BY 4.0 with a link back

CVE-2026-96893 · Severity: high · CVSS 7.3 · Published 2026-10-13

GitLab Enterprise Edition includes an AI agent platform for automating tasks within CI/CD pipelines and project management. An authenticated user could inject malicious content into AI agent conversations, causing the AI to execute unintended commands that appear to be performed by other users. This could lead to unauthorized code changes, pipeline manipulation, or exposure of sensitive project data.

Technologies: [GitLab](https://junglewise.ai/threats/technologies/gitlab), [GitLab Enterprise Edition](https://junglewise.ai/threats/technologies/gitlab-ee). Vendors: [GitLab](https://junglewise.ai/threats/vendors/gitlab).

## Quick answers

**Is CVE-2026-96893 exploited in the wild?** No. Junglewise Threat Intelligence has no report of CVE-2026-96893 being exploited in the wild.

**How severe is CVE-2026-96893?** High severity, with a CVSS score of 7.3, published 13 October 2026.

**Which products and versions are affected?** GitLab Enterprise Edition 19.1 before 19.2.8, 19.3 before 19.3.5, 19.4 before 19.4.2.

**Is there a fix for CVE-2026-96893?** Yes: fixed in GitLab EE 19.2.8, 19.3.5, 19.4.2. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.

**How is CVE-2026-96893 attacked?** It is attacked over the network, in a web application.

**Am I affected by CVE-2026-96893?** Check whether you run GitLab or GitLab Enterprise Edition at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

## How to fix

- Fixed in GitLab EE 19.2.8, 19.3.5, 19.4.2

## Technical details

An improper neutralization vulnerability in AI conversation context construction allows authenticated users to inject specially crafted input into AI agent instructions, leading to prompt injection. The vulnerability affects the Duo Agent Platform component and requires authentication and user interaction. A successful exploit could result in arbitrary command execution attributed to other users without proper authorization checks.

## Affected products

- GitLab GitLab Enterprise Edition 19.1 before 19.2.8, 19.3 before 19.3.5, 19.4 before 19.4.2

## Timeline

- 2026-10-13: disclosed
- 2026-10-12: patched

## References

- [https://docs.gitlab.com/releases/patches/patch-release-gitlab-19-4-2-released/](https://docs.gitlab.com/releases/patches/patch-release-gitlab-19-4-2-released/)
- [https://gitlab.com/gitlab-org/gitlab/-/work_items/607968](https://gitlab.com/gitlab-org/gitlab/-/work_items/607968)

## Related threats

- [GitLab EE incomplete enforcement of SAML/LDAP membership lock](https://junglewise.ai/threats/cve-2026-16255-gitlab-has-remediated-an-issue-in-gitlab-ee-affecting-all)
- [GitLab cache poisoning in OAuth/OIDC discovery](https://junglewise.ai/threats/cve-2026-112212-gitlab-has-remediated-an-issue-in-gitlab-ce-ee-affecting-all)
- [GitLab has remediated an issue in GitLab EE affecting all versions from 13.6 before 19.2.8, 19.3 before 19.3.5, and…](https://junglewise.ai/threats/cve-2026-9268-gitlab-has-remediated-an-issue-in-gitlab-ee-affecting-all-versions)
- [GitLab CE/EE integer overflow in CI/CD regular expression compiler](https://junglewise.ai/threats/cve-2026-97229-gitlab-has-remediated-an-issue-in-gitlab-ce-ee-affecting-all)
- [GitLab Community Edition and Enterprise Edition cross-site scripting in syntax highlighting](https://junglewise.ai/threats/cve-2026-92883-gitlab-has-remediated-an-issue-in-gitlab-ce-ee-affecting-all)