vLLM: VLLM versions >= 0.6.3 and < 0.9.0 contain multiple regular expression denial of service (ReDoS)…
vLLM
More for vLLM RSS feed for vLLM
- Severity
- High, CVSS 7.5
- CVSS vector
- CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H
- Record
- CVE-2025-71379
- Affects
- vLLM
- Known exploited
- Not listed in the CISA Known Exploited Vulnerabilities catalog
- Published
- Record first published by the source
- June 20, 2026
- Source last modified
- October 12, 2026
Description
vLLM versions >= 0.6.3 and < 0.9.0 contain multiple regular expression denial of service (ReDoS) vulnerabilities. Several regex patterns — in vllm/lora/utils.py, the phi4mini tool parser, and the OpenAI-compatible serving chat endpoint — are susceptible to catastrophic backtracking. An attacker submitting crafted input with nested or repeated structures can trigger severe CPU consumption and performance degradation, resulting in denial of service.
Description quoted from the NVD record.
Who is affected
vLLM is software that runs AI language models on a server so that other apps and users can send them questions and get answers. This flaw affects versions 0.6.3 up to but not including 0.9.0. Several parts of the product are involved: the part that handles add-on model adapters, a tool-handling feature for the Phi-4-mini model, and the chat feature that works like OpenAI's chat service. An attacker who sends specially crafted text with nested or repeated patterns can make the server use a huge amount of processing power. This can slow it badly or stop it from serving anyone. This flaw is rated High.
What to do
- Find out who installed or runs vLLM for your organization: a staff member, a volunteer, or an IT provider.
- If you run it yourselves, check which version is installed. Any version from 0.6.3 up to but not including 0.9.0 has this flaw.
- Update to version
0.9.0or later, which is outside the affected range, and check the vendor advisory linked on this page to confirm the fixed release. - If an IT provider or vendor runs vLLM for you, send them this advisory and ask when the update will be done.
Analysis written from the source record as of October 12, 2026 and reviewed before publishing.