Vulnerability in Vllm-Project Vllm
CVE-2026-73557
vLLM is an inference and serving engine for large language models. From 0.20.2rc0 until 0.26.0, safe_load_prompt_embeds in vllm/renderers/embed_utils.py uses torch.sparse.check_sparse_tensor_invariants, whose process-global save, enable, and restore state can be raced by concurrent prompt_embeds parts submitted to POST /v1/chat/completions through AsyncMultiModalItemTracker.resolve_items, asyncio.gather, and the default executor, allowing an invalid sparse tensor to reach tensor.to_dense despite the CVE-2025-62164 guard when enable_prompt_embeds is enabled. This issue is fixed in version 0.26.0.
Vulnerability class: Race Condition
Published · last modified .
EPSS exploit prediction
EPSS: 0.004 (32.1th percentile), scored .
Very low probability of exploitation in the next 30 days; routine patching cadence is appropriate. 32nd percentile — 32.1% of CVEs in the catalogue have a lower EPSS than this one. How to read EPSS.
EPSS over last 30 days · oldest: 0.003 · newest: 0.004 · change: +0.002
Affected products
- Vllm-Project Vllm — versions >= 0.20.2rc0, < 0.26.0