Technical impact
Endpoint authentication is required to access ai‑gateway‑payload‑processing; however, the service does not verify that the authenticated caller is authorized to use the tenant identifier in the request. An attacker who can authenticate can supply another tenant’s model identifier in the request body or header, causing the service to retrieve that tenant’s credentials from a shared store and inject them into the request. Based on the description, it is inferred that the attacker can use the injected credentials to issue requests against an unrelated tenant’s model, thereby consuming that tenant’s large‑language‑model resources and quotas.
Affected Systems
Red Hat OpenShift AI (RHOAI) is affected by this flaw. No specific version information is provided, so any deployment of RHOAI should be evaluated for this vulnerability.
Risk and Exploitability
The vulnerability carries a CVSS score of 9.1, indicating a high risk of exploitation without restrictions. The EPSS score is not available, but the flaw is currently listed as not included in CISA’s KEV catalog. Attackers with valid credentials could exploit this issue by crafting requests that reference another tenant’s model identifier, gaining unauthorized access to that tenant’s LLM resources. The attack vector is remote and requires only a typical authenticated session, making it plausible in multi‑tenant environments where tenant isolation is expected to be enforced.