Red Hat Security Advisory: Red Hat AI Inference Server 3.3.6 (ROCm)

The vulnerability (CVE-2026-41523) exists in vLLM, the inference and serving engine used by Red Hat AI Inference Server. It allows an unauthenticated attacker to achieve arbitrary code execution by loading a malicious HuggingFace model while vLLM operates in Python optimized mode. This affects versions of Red Hat AI Inference Server prior to 0.22.0 that include the vulnerable pooler activation loader component. Exploitation requires the attacker to supply an untrusted model and the server to be running with Python optimization enabled. Red Hat classifies the impact as Important for AI Inference Server and OpenShift AI vLLM serving images, and Moderate for RHEL AI bootc images bundling vLLM. Currently, no official fix or patch is available from Red Hat. Mitigation recommendations focus on configuration and operational controls rather than software updates.

Successful exploitation allows unauthenticated attackers to execute arbitrary code on the affected system running Red Hat AI Inference Server with vLLM in Python optimized mode. This can compromise the server integrity and potentially lead to full system compromise. The impact severity is rated Important (high) for AI Inference Server and OpenShift AI vLLM serving images, and Moderate for RHEL AI bootc images bundling vLLM. There are no known exploits in the wild at this time.

No official patch or fix is currently available. Red Hat recommends the following mitigations: avoid running vLLM with Python optimization enabled; load models only from trusted sources; restrict permissions for model deployment to trusted users; and apply network access controls to limit exposure. These mitigations reduce the risk of exploitation until a fix is released.