- CVSS
- HIGH · 7.5v3.1CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H
- Published
- 2026-09-21
- Weakness
- CWE-248
- Source
- nvd.nist.gov/vuln/detail/CVE-2026-94622
Description
vLLM versions through 0.29.0 contain a denial of service vulnerability in the NIXL connector’s metadata handling for prefill/decode disaggregated deployments. Attackers can send requests with incomplete kv_transfer_params dictionary entries to trigger an uncaught KeyError in EngineCore scheduling, causing the decode engine to terminate and making all routed requests fail until manual restart.
References
- https://github.com/vllm-project/vllm
- https://github.com/vllm-project/vllm/blob/v0.29.0/vllm/distributed/kv_transfer/kv_connector/v1/nixl/metadata.py#L294-L312
- https://github.com/vllm-project/vllm/pull/54807
- https://www.vulncheck.com/advisories/vllm-through-0.29.0-denial-of-service-via-incomplete-nixl-kv-transfer-metadata
How GTK Cyber trains on this
AI security training at GTK Cyber covers the LLM and ML-pipeline vulnerability classes that vulnerabilities like CVE-2026-94622 fall into. Our hands-on courses are taught by Charles Givre and other practitioners who break and defend production AI systems.