Incorrect Type Conversion or Cast Affecting vllm package, versions [0.18.0, 0.20.0)
Do your applications use this vulnerable package?
In a few clicks we can analyze your entire application and see what components are vulnerable in your application, and suggest you quick fixes.
Test your applications- Snyk IDSNYK-PYTHON-VLLM-16439017
- published7 May 2026
- disclosed6 May 2026
- creditYunzez :0
Introduced: 6 May 2026
NewCVE-2026-44223 (opens in a new tab)How to fix?
Upgrade vllm to version 0.20.0 or higher.
Overview
vllm is an A high-throughput and memory-efficient inference and serving engine for LLMs
Affected versions of this package are vulnerable to Incorrect Type Conversion or Cast through the extract_hidden_states speculative decoding. An attacker can cause the server to crash and disrupt service availability by submitting a request containing penalty parameters such as repetition_penalty, frequency_penalty, or presence_penalty.
Note: This is only exploitable if the speculative decoding method is set to extract_hidden_states.
Workaround
This vulnerability can be mitigated by avoiding the use of extract_hidden_states as the speculative decoding method or by filtering out penalty parameters from incoming requests at an API gateway.