Monday 16 November
Global

Your region, kept on this device.

vLLM

Product

Vulnerabilities

21 CVEs in our records, 0 known exploited (on CISA's KEV list).

1Critical
11High
8Medium
1Low
0No score
Source: NVD and CISA KEV

Am I affected?

Checked against NVD affected-version data on our server. Nothing you type is stored.

Affected versions

Source: NVD

Security timeline

  1. Vulnerability reported by CSO OnlineCPython memory-safety flaw lets malicious AI model templates escape sandboxes

Recent news

CSO Online·

CPython memory-safety flaw lets malicious AI model templates escape sandboxes

reported a memory-safety flaw in that lets a malicious chat template in an AI model escape the Jinja sandbox and execute code within inference servers such as vLLM. The demonstration used a backdoored model, and the risk is greatest for hosts that let customers upload their own models. The CPython team does not treat sandbox escapes as its security vulnerability, but the fix was merged on 14 August and backported to the 3.13, 3.14 and 3.15 branches as a regular bug fix.

Last updated 11 November 2026. Facts come from our own stories, NVD and CISA KEV, MITRE ATT&CK and Wikidata, each named beside the fact; the introduction, when there is one, is from Wikipedia. We never show company logos.

Report a mistake