GitLab AI Gateway CVE-2026-90970: ein Sandbox-Ausbruch mit 9,9
GitLab released a critical patch for its AI Gateway, addressing CVE-2026-90970 with a CVSS score of 9.9. The vulnerability could allow authenticated users with access to the Duo Agent Platform to execute arbitrary commands on the AI Gateway. Only self-hosted AI Gateway users needed to act, as GitLab had already patched its hosted gateways. The flaw was discovered by a HackerOne researcher, and there is no evidence of exploitation.
How this was made

The 30-second read
Why it matters
The advisory clarifies that GitLab.com instances are already patched, focusing risk on self‑hosted deployments.
Market read
Security advisory may cause short‑term price pressure for GITLAB, but broader market impact is minimal.
What to watch
Potential downstream impact on third‑party AI model providers integrated via the gateway, and on enterprises' overall AI strategy budgets.
Background
GitLab's AI Gateway routes prompts to large language models; a sandbox escape could allow command execution on the gateway host.
Market effects
Highlights security risks in AI middleware, may spur broader scrutiny of AI‑related SaaS offerings.
Primarily affects North American and European enterprises using self‑hosted GitLab AI Gateway.
Limited to GitLab and its self‑hosted customers; no broader market move expected.
Counterpoint
The patch may be seen as a proactive security measure, potentially reinforcing confidence in GitLab's governance.
Key entities
- companyGitLab
Provider of the AI Gateway and source of the CVE advisory.
- government_agencyCISA
U.S. Cybersecurity and Infrastructure Security Agency that rated the exploit likelihood as "none".


