CVE-2026-73557: CWE-362: Concurrent Execution using Shared Resource with Improper Synchronization ('Race Condition') in vllm-project vllm
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.
AI Analysis
Technical Summary
The vulnerability CVE-2026-73557 in vllm (versions 0.20.2rc0 through 0.25.x) involves a race condition (CWE-362) in the safe_load_prompt_embeds function within vllm/renderers/embed_utils.py. This function uses torch.sparse.check_sparse_tensor_invariants, which relies on a process-global save, enable, and restore state. Concurrent prompt_embeds parts submitted via POST /v1/chat/completions can race through AsyncMultiModalItemTracker.resolve_items, asyncio.gather, and the default executor, causing improper synchronization. This race condition allows an invalid sparse tensor to reach tensor.to_dense despite protections introduced by CVE-2025-62164, when enable_prompt_embeds is enabled. The issue is resolved in vllm version 0.26.0.
Potential Impact
An attacker able to submit concurrent prompt embeddings can exploit this race condition to cause an invalid sparse tensor to be processed by tensor.to_dense, potentially leading to unexpected behavior or denial of service. The CVSS 4.0 score is 6.3 (medium severity), indicating a moderate impact with network attack vector, low attack complexity, no privileges or user interaction required, and limited impact on availability.
Mitigation Recommendations
This vulnerability is fixed in vllm version 0.26.0. Users should upgrade to version 0.26.0 or later to remediate this issue. No official patch or workaround is indicated beyond upgrading. Since this is not a cloud service, remediation is the responsibility of the user.
CVE-2026-73557: CWE-362: Concurrent Execution using Shared Resource with Improper Synchronization ('Race Condition') in vllm-project vllm
Description
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.
CVSS v4.0
Score 6.3medium
Affected software
Run on your own infrastructure? Check whether these packages are installed with threat-finder — our free open-source scanner.
Weaknesses
AI-Powered Analysis
Machine-generated threat intelligence
Technical Analysis
The vulnerability CVE-2026-73557 in vllm (versions 0.20.2rc0 through 0.25.x) involves a race condition (CWE-362) in the safe_load_prompt_embeds function within vllm/renderers/embed_utils.py. This function uses torch.sparse.check_sparse_tensor_invariants, which relies on a process-global save, enable, and restore state. Concurrent prompt_embeds parts submitted via POST /v1/chat/completions can race through AsyncMultiModalItemTracker.resolve_items, asyncio.gather, and the default executor, causing improper synchronization. This race condition allows an invalid sparse tensor to reach tensor.to_dense despite protections introduced by CVE-2025-62164, when enable_prompt_embeds is enabled. The issue is resolved in vllm version 0.26.0.
Potential Impact
An attacker able to submit concurrent prompt embeddings can exploit this race condition to cause an invalid sparse tensor to be processed by tensor.to_dense, potentially leading to unexpected behavior or denial of service. The CVSS 4.0 score is 6.3 (medium severity), indicating a moderate impact with network attack vector, low attack complexity, no privileges or user interaction required, and limited impact on availability.
Mitigation Recommendations
This vulnerability is fixed in vllm version 0.26.0. Users should upgrade to version 0.26.0 or later to remediate this issue. No official patch or workaround is indicated beyond upgrading. Since this is not a cloud service, remediation is the responsibility of the user.
Technical Details
- Data Version
- 5.2
- Assigner Short Name
- GitHub_M
- Date Reserved
- 2026-08-12T20:53:46.380Z
- Cvss Version
- 4.0
- State
- PUBLISHED
- Remediation Level
- null
Threat ID: 6a7ddecdbf8831d5395d0186
Added to database: 08/13/2026, 15:12:13 UTC
Last enriched: 08/13/2026, 15:28:15 UTC
Last updated: 08/13/2026, 17:43:31 UTC
Views: 3
Community Reviews
0 reviewsCrowdsource mitigation strategies, share intel context, and vote on the most helpful responses. Sign in to add your voice and help keep defenders ahead.
Want to contribute mitigation steps or threat intel context? Sign in or create an account to join the community discussion.
Actions
Updates to AI analysis require Pro Console access. Upgrade inside Console → Billing.
Need more coverage?
Upgrade to Pro Console for AI refresh and higher limits.
For incident response and remediation, OffSeq services can help resolve threats faster.
Latest Threats
Check if your credentials are on the dark web
Instant breach scanning across billions of leaked records. Free tier available.