Skip to main content
All incident reports
Incident report

GitLab patches critical AI Gateway flaw allowing command execution

GitLab has released fixes for a critical vulnerability in its self-hosted AI Gateway, tracked as CVE-2026-90970, that let an authenticated user with Duo Agent Platform access run commands on the gateway. Customers using a GitLab-hosted gateway are already protected and need do nothing, the company says.

· First published

The short version

  • The flaw. CVE-2026-90970 is a critical weakness in the GitLab AI Gateway, the service that connects a GitLab instance to AI models. A logged-in user with Duo Agent Platform access could use a crafted flow configuration to break out of the prompt template sandbox and execute commands on the gateway. The Hacker News reports a CVSS score of 9.9 out of 10.
  • Who must act. Only organisations running their own self-hosted AI Gateway need to patch. GitLab operates the gateway for GitLab.com, GitLab Dedicated and self-managed instances that use the hosted service, and says those are already fixed. GitLab contacted self-hosted gateway customers before publishing the advisory.
  • Older releases. The fixed gateway versions are 19.2.4, 19.3.2 and 19.4.1. No fixed version is listed below 19.2.4, so every gateway release from 18.1.6 through the 19.1 line sits in the affected range with no workaround, according to The Hacker News.
  • Exploitation status. GitLab's advisory does not say whether the flaw has been used in attacks. CISA added an assessment to the CVE record on 2 October listing exploitation as none, and no source reports an attack.
  • Earlier GitLab flaws. GitLab fixed a gateway flaw of the same weakness class, CVE-2026-1868, in February, also rated 9.9. It had also patched a maximum-severity path traversal flaw, CVE-2026-85706, which CISA added to its known exploited vulnerabilities catalogue the day after the patch.

Timeline

DateEvent
FebruaryGitLab fixes an earlier AI Gateway flaw, CVE-2026-1868, rated 9.9 and of the same template engine weakness class as the October flaw.
Before 2 OctoberBefore publication on 2 October: GitLab contacts customers that host their own AI Gateway with patching guidance ahead of publication.
2 OctoberGitLab publishes its advisory for CVE-2026-90970 and releases fixed AI Gateway versions; The Hacker News reports a CVSS score of 9.9.
2 OctoberCISA adds an assessment to the CVE record listing exploitation as none.

What went wrong and what fixes it

GitLab published an advisory on 2 October for CVE-2026-90970, a critical weakness in the AI Gateway, the component that links a GitLab instance to AI model providers. According to GitLab, an authenticated user with Duo Agent Platform access could use a specially crafted flow configuration to "escape the prompt template sandbox" and run commands on the gateway. The Hacker News reports a CVSS score of 9.9 out of 10.

The weakness sits in the prompt template of a custom flow, an AI workflow that users build on the Duo Agent Platform to automate multi-step tasks. GitLab credited the HackerOne user invisiblemeerkat with reporting it. The advisory does not describe the conditions an attack requires or name any role beyond Duo Agent Platform access. An earlier gateway flaw fixed in February, CVE-2026-1868, was of the same template engine class, CWE-1336, and also rated 9.9; the October advisory does not mention it.

Who is affected

Only organisations that host their own AI Gateway need to act. GitLab runs the gateway for GitLab.com, GitLab Dedicated and self-managed instances that use the hosted service, and says those are already protected. Self-managed customers who run GitLab Duo Self-Hosted operate their own gateway, an option GitLab offers to keep AI request and response data inside the customer's environment. GitLab says it reached out to those customers before publication.

The fixed gateway releases are 19.2.4, 19.3.2 and 19.4.1. The Hacker News notes that no fixed version is listed below 19.2.4, leaving releases from 18.1.6 through the 19.1 line in the affected range, and that the advisory gives no workaround and no way to check whether a gateway was attacked before it was updated. It also does not say whether a 19.2.4 gateway works with GitLab 19.1 or earlier. A self-hosted gateway holds JSON Web Token signing keys, which GitLab's install guide says must be treated as sensitive credentials, and connects to the GitLab instance and the organisation's AI model providers.

Exploitation and wider context

GitLab's advisory does not say whether the flaw has been exploited. CISA added an assessment to the CVE record on 2 October listing exploitation as "none", one of three values covering no known exploitation, a public proof of concept and active exploitation. No source reports an attack against a gateway.

GitLab has a record of exploited flaws. A maximum-severity path traversal flaw that it patched before this disclosure, CVE-2026-85706, was added by CISA to its known exploited vulnerabilities catalogue the day after the patch, with federal agencies given three days to secure their systems. BleepingComputer reports that CISA has tagged five GitLab vulnerabilities as abused in the wild since November 2021, including one exploited by ransomware gangs. GitLab's platform has more than 30 million registered users and is used by over half of Fortune 100 companies, the same outlet reports.

What is not known yet

  • Whether anyone exploited CVE-2026-90970 before self-hosted gateways were updated, and how an administrator could tell.
  • Who, if anyone, carried out an attack; no source names an actor or reports one.
  • How many self-hosted AI Gateway deployments exist or how many are exposed.
  • Whether GitLab will issue fixes for gateway releases 18.1.6 through the 19.1 line, which have no listed fixed version.
  • Whether a 19.2.4 gateway is compatible with GitLab 19.1 or earlier, or a fix for older lines is planned.

What to do

If you run a self-hosted GitLab AI Gateway

  1. Update the gateway to 19.4.1; GitLab's advisory lists the fixed release for each supported line and the matching update steps.
  2. For a Docker deployment, stop and remove the running container, then pull and run the new image tag; for Helm, set the new tag in the chart's image setting.
  3. Check your current version against the advisory before relying on an upgrade if you run 18.1.6 through the 19.1 line, where no fixed version is listed.
  4. Treat the gateway's JSON Web Token signing keys as sensitive credentials, as GitLab's install guide requires.

If you use GitLab.com, GitLab Dedicated or a hosted gateway

  1. Confirm whether your instance uses the GitLab-hosted gateway or a self-hosted one; only the self-hosted case needs patching.
  2. Take no upgrade action on the hosted gateway, which GitLab says is already protected.

For everyone else

  1. Review who holds Duo Agent Platform access, since the flaw needed only an authenticated user with that access rather than administrative rights.
  2. Put self-hosted AI components such as gateways on the same patch cycle as the main platform, rather than treating them as add-ons.

Update log

  • First published.

Sources

New incident reports by email

One email a week with the reports published since the last one.

Want us to look at your domain?

Give us one apex domain and we'll send back a written snapshot of what is already visible from outside. No call required.