CVE-2026-54236
MediumCVSS 5.3Exploitation Probability (EPSS)
Elevated risk60th percentile - higher than 60% of all known CVEs
Summary
vLLM is an inference and serving engine for large language models. In versions prior to 0.23.1rc0, an incomplete fix for CVE-2026-22778 allows memory addresses to leak in error messages, potentially exposing memory information in responses to clients.
Risk Assessment
The organization may be vulnerable to attacks where unauthorized users can access sensitive memory information, leading to further exploits or attacks on the system.
Recommendation
It is recommended to upgrade to version 0.23.1rc0 or later to patch this vulnerability and secure the system against potential attacks.
Related vulnerabilities
- CVE-2026-113364Critical
Buffer overflow in the data receiving process in NEC WebSAM WinShare versions 7.5.0 and earlier, 7.6.0, and 7.7.0 through 7.7.2 on Windows allows a remote unauthenticated attacker to execute arbitrary code with SYSTEM privileges via a crafted packet with a malformed data size value in the protocol header.
- CVE-2026-113363Critical
Improper input validation in the network communication processing in NEC WebSAM WinShare versions 7.2.0 through 7.5.0, 7.6.0, and 7.7.0 through 7.7.2 on Windows allows a remote unauthenticated attacker to execute arbitrary commands with SYSTEM privileges via a crafted network packet sent to the computer running WinShare.
- CVE-2026-124560Critical
A vulnerability was found in D-Link DIR-878 version 1.20B05 in the GetCAPTCHAsetting function of bin/prog.cgi (HNAP Handler component). Manipulating the uid argument results in a heap-based buffer overflow. The attack can be initiated remotely and the exploit has been made public.
- CVE-2026-119506Critical
The Frontend Admin by DynamiApps plugin (acf-frontend-form-element) through version 3.30.1 contains an incorrect privilege assignment vulnerability. The flaw allows an attacker to escalate privileges within the WordPress site.
- CVE-2026-6715Critical
IBM Concert 1.0 through 3.0.1.1 allows an attacker to supply crafted input that is incorporated into an operating system command without adequate validation or sanitization. Successful exploitation allows remote attackers to execute arbitrary commands on the underlying operating system with the privileges of the affected service.
- CVE-2026-17466Critical
IBM Optim versions 1.0.0, 1.1.0, and 1.2.0 are vulnerable to path traversal, allowing a remote authenticated attacker to obtain sensitive information.
- CVE-2026-15266Critical
IBM Concert 1.0 through 3.0.1.1 contains components that allow potentially unsafe command execution, which could be exploited to run arbitrary system commands on the host.
- CVE-2026-114464Critical
IBM Process Mining 2.0.0 through 2.2.1 has improper access control, which could allow a remote attacker to obtain sensitive information.
- CVE-2026-107331Critical
IBM DataStage on Cloud Pak for Data 5.4.0.0 has an absolute-path traversal vulnerability that could allow a remote authenticated attacker to obtain sensitive information.
- CVE-2026-109900Critical
The Melapress Role Editor plugin for WordPress (through version 1.3.0) contains an incorrect privilege assignment vulnerability that allows privilege escalation. A low-privileged user can gain a higher level of access within the system.
Original NVD description (English source)
vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.23.1rc0, the fix for CVE-2026-22778, which introduced a sanitize_message helper that strips object-repr memory addresses from error messages before they reach the client, is incomplete: several response paths echo str(exc) directly to clients without calling sanitize_message. The unsanitized sites include the Anthropic API router in vllm/entrypoints/anthropic/api_router.py (the POST /v1/messages and POST /v1/messages/count_tokens handlers), the Server-Sent Events streaming converter in vllm/entrypoints/anthropic/serving.py, and the realtime speech-to-text WebSocket in vllm/entrypoints/speech_to_text/realtime/connection.py. These paths catch the exception inside the route coroutine and construct the JSONResponse themselves, bypassing the sanitizing global FastAPI exception handler, and WebSocket frames do not traverse that handler chain at all. Using the same primitive as the parent issue, an unauthenticated attacker can send malformed image bytes through the Anthropic Messages API image content parts so that PIL.Image.open raises an UnidentifiedImageError whose message contains the BytesIO object repr, leaking the heap memory address verbatim in the error.message field of the response body. This vulnerability is fixed in 0.23.1rc0.
Vulnerability data from NVD (NIST) · CISA KEV · EPSS