Bug 2515401 (CVE-2026-73557) - CVE-2026-73557 vllm: vLLM: Incomplete remediation allows bypass via concurrent prompt processing
Summary: CVE-2026-73557 vllm: vLLM: Incomplete remediation allows bypass via concurren...
Keywords:
Status: NEW
Alias: CVE-2026-73557
Product: Security Response
Classification: Other
Component: vulnerability
Version: unspecified
Hardware: All
OS: Linux
medium
medium
Target Milestone: ---
Assignee: Product Security
QA Contact:
URL:
Whiteboard:
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2026-08-13 15:51 UTC by OSIDB Bzimport
Modified: 2026-08-19 16:14 UTC (History)
5 users (show)

Fixed In Version:
Clone Of:
Environment:
Last Closed:
Embargoed:


Attachments (Terms of Use)

Description OSIDB Bzimport 2026-08-13 15:51:29 UTC
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.


Note You need to log in before you can comment on or make changes to this bug.