High [CVE-2026-54234] Denial of Service via malformed speculative decoding workload
This high-severity Red Hat Linux advisory covers CVE-2026-54234 affecting Red Hat AI Inference Server, Red Hat Enterprise Linux AI (RHEL AI) 3, Red Hat OpenShift AI (RHOAI).
Android app · Google Play
Monitor future Red Hat Linux CVEs from your phone.
Choose a whole vendor or a precise platform, then receive matching security advisories by phone notification, email, or both. Coverage follows 32 official vendor sources and 160+ reviewed platform categories.
Summary
vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs.
Prior to 0.24.0, a frontend-legal multi-request speculative decoding workload can cause the rejection sampler to produce a recovered token equal to the model vocabulary size boundary value, which is then converted to negative one when the engine selects the next live token for a request and is written back into the drafter's input ids; that out-of-vocabulary value is later consumed by the model's embedding and attention path and crashes the engine worker with a GPU device-side assertion.
The same triggering request sequence is reachable through the public gRPC Generate and Abort endpoints, so a remote client that can send generation requests can crash the shared engine worker, aborting concurrent requests and causing a service-wide denial of service for other clients of the deployment until the worker is restarted.
This issue is fixed in version 0.24.0. A remote attacker can exploit this vulnerability by sending a specially crafted multi-request speculative decoding workload through public gRPC Generate and Abort endpoints.
This Important denial of service flaw in vLLM, utilized by Red Hat AI Inference Server, Red Hat Enterprise Linux AI, and Red Hat OpenShift AI, allows a remote attacker to crash the engine worker.
- < 0.24.0
Official advisory · high-confidence parse· fetched 13 days ago·verify at source
Mitigation checklist
- To mitigate this issue, restrict network access to the vLLM inference engine's gRPC Generate and Abort endpoints. Configure firewall rules to limit incoming connections to trusted clients or internal networks only. This will prevent remote, unauthenticated attackers from sending malformed workloads and triggering a denial of service. If the service is exposed via a proxy or load balancer, ensure that access controls are in place at that layer.
Official advisory · high-confidence parse· fetched 13 days ago·verify at source
Discussion(0)
No comments yet. Share field notes, upgrade gotchas, or questions — verify against the vendor advisory before acting on community advice.
Sign in to join the discussion.