CVE-2026-94626: vLLM through 0.29.0 Memory Exhaustion via Unvalidated NIXL tp_size
vLLM through 0.29.0 fails to validate the tpsize parameter in kvtransferparams on OpenAI-compatible completion endpoints, allowing attackers to allocate unbounded memory. Attackers can supply arbitrary tpsize values in prefill/decode disaggregated deployments to exhaust memory and trigger kernel OOM-kill of the decode worker process.
Affected Software
Event History
Frequently Asked Questions
Which deployments are exposed to this issue?
The issue applies to vLLM deployments through 0.29.0 that use prefill/decode disaggregation and accept OpenAI-compatible completion requests. The affected path processes kv_transfer_params with the NIXL tp_size parameter.
What does an attacker need to exploit it?
An attacker needs network access to an affected OpenAI-compatible completion endpoint. No authentication, user interaction, or special exploit conditions are indicated; they can submit an arbitrary tp_size value in kv_transfer_params.
What is the practical impact of exploitation?
A malicious tp_size value can cause unbounded memory allocation, exhausting memory on the decode worker. This can trigger the kernel OOM killer and terminate the decode worker process, resulting in denial of service.