Reachable Assertion Affecting vllm package, versions [,0.28.0)
Threat Intelligence
The probability is the direct output of the EPSS model, and conveys an overall sense of the threat of exploitation in the wild. The percentile measures the EPSS probability relative to all known EPSS scores. Note: This data is updated daily, relying on the latest available EPSS model version. Check out the EPSS documentation for more details.
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-20877491
- published13 Oct 2026
- disclosed6 Oct 2026
- creditUnknown
Introduced: 6 Oct 2026
NewCVE-2026-105753 (opens in a new tab)How to fix?
Upgrade vllm to version 0.28.0 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 Reachable Assertion via a P0/P1 multimodal processor cache drift condition in vllm/multimodal/cache.py and the V1 engine core. When the frontend (P0) and the engine (P1) multimodal processor caches diverge - because the P0 sender cache is populated before admission checks complete or before P1 acknowledges the item - the engine receives a request referencing a multimodal hash it has already evicted, while the frontend sends no inline data for that item. This mismatch causes the engine to be unable to process the request, resulting in a crash that denies service to all users of the affected instance. An authenticated remote attacker with access to submit inference requests can trigger this condition by sending multimodal requests that cause cache eviction and re-submission races.