GPU VulnDB

Database/Control plane, storage & DevOps

GitLab EE: injected AI conversation context causes agent commands to run as another user

CVSS 7.3CVE-2026-96893Control plane, storage & DevOpscurated

Impact

An authenticated GitLab user can plant crafted content that GitLab's AI agent later folds into its conversation context without neutralizing special elements, so agent-driven command execution can be attributed to a different user. Where GitLab is the CI/CD control plane for a GPU fleet, agent actions carry pipeline and repository privileges - the path from injected text to a job that runs on build or GPU runners is the thing to worry about, not the chat output itself. GitLab's note is deliberately brief; the exact reachable command surface is not detailed in the record, and the score (confidentiality and integrity high, availability none, user interaction required) is all that is established.

Who can reach it

Any authenticated GitLab user who can write content that lands in another user's AI conversation context. Victim interaction with the AI feature is required.

What to do

Upgrade GitLab EE to 19.2.8, 19.3.5 or 19.4.2 and restart the application - a GitLab-side upgrade, no fleet maintenance window. If the upgrade cannot be scheduled immediately, consider disabling AI agent features for untrusted-contributor projects.

References

Related entries

All Control plane, storage & DevOps entries

This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.