GitLab warns of critical RCE vulnerability in AI Gateway service

GitLab has issued a warning and released patches for a critical vulnerability identified as CVE-2026-90970. This flaw exists within the company's AI Gateway service and is categorized as a remote code execution (RCE) vulnerability. According to coverage, the security flaw allows for command execution on servers that are self-hosted. The vulnerability is particularly severe because it allows unauthenticated users to perform data exfiltration, meaning attackers do not need valid credentials to access or remove sensitive information from the affected systems. Multiple cybersecurity and tech news outlets have emphasized the severity of this event. BleepingComputer reports on the warning issued by GitLab regarding the AI Gateway service, while The Hacker News highlights that the flaw enables command execution specifically on self-hosted servers.

InfoQ reports that the vulnerability was under active exploitation at the time of reporting, focusing on the risk of unauthenticated data exfiltration. Additionally, reports from forkast.news and Pasquale Pillitteri confirm that GitLab has officially released patches to address the security gap. The technical stakes of this vulnerability are reflected in its score, which is listed as a 9.9. This high rating indicates a critical level of risk to the infrastructure of organizations utilizing the AI Gateway. The context provided by the coverage suggests that the primary target of this vulnerability is the self-hosted environment, where administrators are responsible for their own server maintenance and patching schedules. The ability for an unauthenticated actor to execute commands and exfiltrate data makes this a high-priority security event for GitLab users.

Moving forward, the focus remains on the deployment of the available patches. Because coverage from InfoQ indicates the flaw was under active exploitation, the immediate priority for affected organizations is to apply the updates released by GitLab. The situation highlights the ongoing risks associated with AI-integrated services and the necessity of rapid patching cycles for self-hosted enterprise software. Further updates will likely depend on the rate of adoption of these patches across self-hosted GitLab installations.