CVE-2026-73557 vllm-project CVE debrief

Who should care

Defenders responsible for large language model serving engines, particularly those using vLLM, should assess exposure and apply the fix to prevent potential issues related to invalid sparse tensor processing. This includes verifying if their vLLM deployment is vulnerable and prioritizing assessing exposure and applying the fix. Security teams and operators managing vLLM deployments should review the vulnerability details and take necessary actions to remem

Why it matters

CVE-2026-73557 is a medium-severity vulnerability in the vLLM inference and serving engine for large language models. Defenders should assess exposure and apply the fix to prevent potential issues.

• Defenders should verify if their vLLM deployment is vulnerable to invalid sparse tensor processing
• Defenders should prioritize assessing exposure and applying the fix to prevent potential issues

Technical summary

The vulnerability arises from a race condition in the safe_load_prompt_embeds function, which can be triggered through concurrent submissions to the POST /v1/chat/completions endpoint. The issue is fixed in version 0.26.0. The vulnerability allows an invalid sparse tensor to reach tensor.to_dense despite the CVE-2025-62164 guard when enable_prompt_embeds is enabled. This issue impacts defenders using vLLM for large language model serving engines, particularly those with exposure to concurrent submissions to the POST /v1/chat/completions endpoint.

Defensive priority

Defenders should prioritize assessing exposure and applying the fix, as the vulnerability can lead to invalid sparse tensor processing.

Recommended defensive actions

• Assess exposure to the vulnerable version of vLLM
• Apply the fix in version 0.26.0
• Monitor for concurrent submissions to the POST /v1/chat/completions endpoint
• Review compensating controls for exposed systems while remediation is scheduled and verified
• Check relevant monitoring, detection, and logs for exposed assets that need extra review
• Track exceptions, retest remediated assets, and close the item only after evidence is documented
• Confirm whether affected product deployments exist in managed environments and assign an owner for follow-up

Evidence notes

The vulnerability is caused by a race condition in the safe_load_prompt_embeds function, which can be triggered through concurrent submissions to the POST /v1/chat/completions endpoint. The issue is fixed in version 0.26.0.

Sources and references

Methodology and review provenance

AI-assisted synthesis based on stored public vulnerability evidence. System validation, approval state, and publication status do not by themselves establish human review of this revision. PatchSiren helps prioritize defensive review and does not prove exposure or remediation on any system.