Opening Forensia workspaceacquiring
Loading the current evidence view. Navigation and account controls remain available.
Loading recent evidenceacquiring
Indicator investigation
Loading the current evidence view. Navigation and account controls remain available.
Loading verdict, provenance, relationships and sightings.
Indicator
Type cve · source NVD CVE
Grouped attempts with latest status; expand for attempt history.
25 attempts · first 2026-07-07T19:58:24.760517Z · last 2026-09-05T19:30:54.903820Z
Not listed in CISA KEV (local)
Types: cisa_kev
25 attempts · first 2026-07-07T19:58:24.762222Z · last 2026-09-05T19:30:54.934587Z
vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.22.0, an assert-based security check in vLLM's activation function loading allows any unauthenticated attacker to achieve arbitrary code execution on the server by publishing a malicious HuggingFace model, when vLLM runs in Python optimized mode (python -O or PYTHONOPTIMIZE=1). This vulnerability is fixed in 0.22.
Types: nvd_local
High-signal pivots without leaving the thread you started in search.
Structured pivots from local graph context.
Full investigation canvas with neighbor expansion.