Skip to main content
Mallory
CriticalVulnerability

CVE-2026-90970: Authenticated command execution via prompt-template sandbox escape in GitLab AI Gateway

Published Oct 2, 2026 · Updated 18 days ago
Trending this weekNot trendingOutside this week’s top 100
Mention activity · 30 days1 mention · peak 1 in a dayToday
CVSS9.9Critical · v3.1
EPSS17.5%97th percentile
CISA KEVNoNot listed
In the wildNoneNo exploitation reported
Vendor
GitLab
IdentifiersCVE-2026-90970CWE-1336 · Improper Neutralization of Special…

CVE-2026-90970 is a template-engine vulnerability in GitLab AI Gateway's processing of Duo Agent Platform custom-flow prompt templates. Improper neutralization of template special elements allows an authenticated user with Duo Agent Platform access, under certain conditions, to escape the prompt-template sandbox through a specially crafted flow configuration and execute arbitrary commands on the gateway. Affected versions are 18.1.6 up to but excluding 19.2.4, 19.3 up to but excluding 19.3.2, and 19.4 up to but excluding 19.4.1. The precise vulnerable code path, template engine, and additional exploitation conditions have not been publicly disclosed. The vulnerability has a CVSS v3.1 score of 9.9.

Do you run this?See affected versions in your environment and whether adversaries are exploiting this vulnerability against assets like yours.
Check your exposure
Analyst brief

Impact, mitigation & remediation

What it means and what to do now: patch path, interim mitigations, and the assume-compromise checklist.

Impact

What an attacker gets, and what they’ve been doing with it.

Successful exploitation bypasses the prompt-template security boundary and permits arbitrary command execution in the AI Gateway service context, with potentially high confidentiality, integrity, and availability impacts. Depending on runtime permissions, mounted resources, and network access, compromise could expose AI request data, source code, JWT signing material, model-provider credentials, and other gateway-accessible secrets. It could also enable modification or disruption of gateway operations and access to connected internal services. Host-wide or downstream compromise depends on deployment privileges and accessible resources.

Mitigation

If you can’t patch tonight, do this now.

No vendor-provided workaround is identified. Until patching is complete, restrict Duo Agent Platform and custom-flow access to necessary trusted users, limit gateway network access to required GitLab services, approved model providers, and trusted administrative networks, and restrict unnecessary outbound traffic. These measures reduce exposure but do not eliminate the vulnerability. Review flow-configuration changes, unexpected gateway child processes, suspicious access to mounted files or configuration data, and unusual outbound connections. If command execution is suspected, investigate accessible secrets and rotate potentially exposed credentials. Monitoring and hardening do not replace upgrading.

Remediation

Patch, then assume compromise.

Immediately upgrade affected self-hosted GitLab AI Gateway deployments to 19.2.4, 19.3.2, 19.4.1, or a later applicable fixed release. No fixed releases below 19.2.4 are identified; affected older branches must move to an applicable supported fixed release, with GitLab compatibility requirements checked before migration. Update the AI Gateway independently of the main GitLab application by replacing its Docker image or updating its Helm deployment image. Verify the running image version and digest, and ensure image caching has not retained a vulnerable build. GitLab-hosted AI Gateways have already been patched; customers using those gateways require no action for this vulnerability.

Public exploits

Exploits

No validated public exploits yet. Hidden: 1 rejected by Mallory's analysis as not a working exploit.

View more in app →
Every candidate exploit was rejected by Mallory's analysis.
Exposure surface

Affected products & vendors

Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.

VendorProductType
GitLabAi-Gatewayapplication
FAQ

Frequently asked questions

Is CVE-2026-90970 exploited in the wild?
As of Nov 3, 2026, Mallory has no reports of CVE-2026-90970 being exploited in the wild. It is not listed in CISA KEV.
Is there a public exploit for CVE-2026-90970?
As of Nov 3, 2026, Mallory has found no public exploit for CVE-2026-90970.
Is there a patch for CVE-2026-90970?
Immediately upgrade affected self-hosted GitLab AI Gateway deployments to 19.2.4, 19.3.2, 19.4.1, or a later applicable fixed release. No fixed releases below 19.2.4 are identified; affected older branches must move to an applicable supported fixed release, with GitLab compatibility requirements checked before migration.
What products are affected by CVE-2026-90970?
GitLab Ai-Gateway.
How severe is CVE-2026-90970?
CVE-2026-90970 has a CVSS v3.1 score of 9.9 (Critical) and an EPSS score of 17.5% (97th percentile) as of Nov 3, 2026.
Activity feed

Recent activity

81 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.

View more in app →
krcert kisaNews · Oct 14, 2026GitLab 제품 보안 업데이트 권고 > 보안공지 > 알림마당 : KISA 보호나라&KrCERT/CCAn improper neutralization of special elements used in a template engine vulnerability in GitLab AI Gateway. Affected versions are 18.1.6 through versions before 19.2.4, 19.3 through versions before 19.3.2, and 19.4 through versions before 19.4.1. The advisory recommends updating to the corresponding fixed release or later.Read more ↗xakepNews · Oct 5, 2026В GitLab AI Gateway исправили критический баг на 9,9 балла - ХакерA critical vulnerability rated CVSS 9.9 in self-hosted GitLab AI Gateway allows an authenticated user with Duo Agent Platform access to escape the prompt-template sandbox through a specially crafted flow configuration and execute arbitrary commands on the gateway. The minimum required privileges beyond platform access are not disclosed. Affected versions begin at 18.1.6; fixes are available in 19.2.4, 19.3.2, and 19.4.1. Versions 18.1.6 through 19.1 remain vulnerable without fixes on those branches. Compromise could expose JWT signing keys and affect interactions with GitLab instances and AI-model providers. AI Gateway must be updated independently of GitLab. GitLab-hosted gateways are already protected and require no customer action.Read more ↗socradar blogNews · Oct 5, 2026CVE-2026-90970: GitLab AI Gateway RCEA critical template-engine vulnerability (CWE-1336) in GitLab's Self-Hosted AI Gateway allows an authenticated user with Duo Agent Platform access to submit a crafted flow configuration, escape the prompt-template sandbox, and execute arbitrary commands. Affected versions are 18.1.6 through versions before 19.2.4, 19.3 before 19.3.2, and 19.4 before 19.4.1. Fixed releases are 19.2.4, 19.3.2, and 19.4.1. GitLab-hosted gateways have already been patched. Recommended mitigations include restricting Duo Agent Platform and custom-flow access and limiting gateway network access; these do not replace patching.Read more ↗mkd cirtNews · Oct 5, 2026GitLab ја поправа критичната ранливост 9.9 во AI Gateway, која овозможува извршување команди на самохостирани сервери | MKD-CIRT | Национален центар за одговор на компјутерски инцидентиA critical custom-flow template sandbox escape in GitLab AI Gateway, rated CVSS 9.9. Exploitation requires authentication and access to Duo Agent Platform; additional necessary conditions are unspecified. Affected Gateway versions are 18.1.6 through versions before 19.2.4, 19.3 before 19.3.2, and 19.4 before 19.4.1. Fixes are available in 19.2.4, 19.3.2, and 19.4.1. GitLab-hosted gateways have already been updated; organizations operating their own gateways must update them. The advisory does not disclose exploitation status, while CISA's October 2 assessment lists exploitation as 'none'. No temporary workaround or method for checking prior compromise is provided.Read more ↗
+10 more in the timeline. Sign up to see them
Do you run this?See affected versions in your environment and whether adversaries are exploiting this vulnerability against assets like yours.
Check your exposure
Mallory platform

The version that knows your environment.

This page is what’s public. Mallory adds the parts that aren’t: which of your assets are affected, which adversaries are exploiting it right now, and which detections to deploy tonight.
Exposure mappingQuery your assets running an affected version, and investigate the blast radius.
Threat actor evidenceEvery report linking this CVE to a named adversary.
Associated malwareMalware families linked to this CVE, with evidence and IOCs.
Detection signaturesYARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Vendor-by-vendor mappingCross-references every affected SKU, including bundled OEM variants.
Social activity65Community discussion across Reddit, Mastodon, and other social sources.