CVE-2026-92220: vLLM MoRIIO resource exhaustion in acknowledgement handler

Executive brief

vLLM is a high-throughput inference engine for large language models. A vulnerability in the MoRIIO acknowledgement handler allows remote attackers to manipulate parameters and trigger excessive resource consumption, potentially causing service degradation or denial of service.

Technical details

The vulnerability exists in the MoRIIO Acknowledgement Handler component (vllm/distributed/kv_transfer/kv_connector/v1/moriio/moriio_connector.py), specifically in the MoRIIOConnectorScheduler.request_finished, MoRIIOConnectorWorker.get_finished, and MoRIIOWrapper._handle_release_message functions. By manipulating the request_id or kv_transfer_params arguments, an attacker can cause uncontrolled resource consumption without requiring authentication. The attack is remotely exploitable. A fix has been proposed via pull request #50674 to reject duplicate release messages for already-terminal transfers.

Affected products

• vLLM Project vLLM 0.26.0, 0.27.0

Timeline

• 2026-09-16: disclosed
• 2026-08-01: other: Fix proposed via pull request #50674

References

• https://github.com/vllm-project/vllm/
• https://github.com/vllm-project/vllm/pull/50674