The flaw in the vllm-orchestrator-gateway component causes the production binary to log every incoming authorization header and the full chat payload. This weakness, classified as CWE-538, may contain personally identifiable information and secrets, such as bearer tokens and conversation content. Because the logs are persistent, any user with access to the logs can read this sensitive information, leading to the disclosure of credentials and private content.

This vulnerability affects the vllm-orchestrator-gateway component of Red Hat OpenShift AI (RHOAI). All deployments of this component are potentially impacted; no specific version range is listed, so the risk applies to all current releases unless patched.

The CVSS score of 7.5 indicates a moderate to high severity. The EPSS score is less than 1 %, indicating a low likelihood exploitation at this time. The vulnerability is not listed in CISA’s KEV catalog. Because the logs are accessible to anyone with logging privileges, an attacker who gains or already has such privileges can read tokens and content; therefore the attack vector is likely internal or through compromised monitoring accounts. No active exploit is reported, but the information leaking could facilitate credential theft or lateral movement.