GPU VulnDB

Database/Control plane, storage & DevOps

Grafana: unsanitized alert generatorURL runs attacker JavaScript in a viewing user's session

CVE-2026-17033Control plane, storage & DevOpscurated

Impact

A user with Editor access or alert.instances.external:write can submit an external Alertmanager alert whose generatorURL carries a script payload; Grafana renders it as the 'See source' link href with no scheme allowlist, and the click interceptor's :// heuristic is bypassed by hiding :// inside a JavaScript comment. When any user with read access clicks that link, the script runs in the Grafana origin with the clicking user's permissions - so an Editor escalates to whatever the victim can do, up to admin actions like editing data sources or minting tokens if an admin clicks. On a GPU fleet Grafana usually fronts DCGM and fabric telemetry and holds credentials into the rest of the monitoring stack. Exploitation depends on a victim click, so this is not self-propagating.

Who can reach it

Authenticated Grafana user holding the Editor role or alert.instances.external:write, over the network. The payload only fires when a different user opens Alert Details and clicks 'See source'.

What to do

Upgrade Grafana to the release named in Grafana's advisory for CVE-2026-17033 and restart the Grafana server; the NVD record carries no fixed version. Grafana is a single service, so this is a daemon restart, not a fleet action - no GPU node is drained or rebooted. In the interim, review who holds Editor and external alert-write permissions, since the flaw needs one of those roles to plant the alert.

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.