CVE intelligence and bounded remediation

CVE-2026-48746 — vLLM is an inference and serving engine for large language models (LLMs)

Critical CVSS 9.1

vLLM is an inference and serving engine for large language models (LLMs). From 0.3.0 until 0.22.0, a vulnerability in ASGI web servers and starlette's trust on those web servers enables an authentication bypass of the OpenAI API AuthenticationMiddleware. It allows to use the API without providing the configured VLLM_API_KEY or --api-key. This vulnerability is fixed in 0.22.0.

Severity
Critical
CVSS
9.1 (3.1)
Published
2026-06-22
CISA KEV
Not currently listed
Ecosystem
software/application
Weaknesses
CWE-444, CWE-501

Affected products

  • vllm / vllm

Matched remediation archetype

HTTP request smuggling and message-boundary ambiguity

This catalog composition supplies bounded fallback guidance. Explicitly reviewed curated workflows load with the complete record below.

Check exposure

  • Inventory every proxy, CDN, gateway, load balancer, service mesh, and application server hop on affected request paths.
  • Compare documented parsing and normalization behavior for message length, transfer coding, duplicate headers, and protocol translation.
  • Identify connection reuse and which downstream services trust headers added by intermediaries.

Remediate safely

  • Update affected intermediaries and origin servers and align them on a single standards-compliant request framing policy.
  • Reject ambiguous length, transfer-coding, duplicate, malformed, and unsupported framing before forwarding.
  • Normalize or remove hop-by-hop and identity headers at one controlled boundary and add multi-hop regression tests.

Authoritative sources

Complete CVE record and remediation plan

The detailed catalog view below loads this exact record, its source evidence, and the full seven-phase agentic change plan.