CVE-2026-55574 vllm-project CVE debrief

Who should care

Users of vLLM inference and serving engine for LLMs, particularly those who manage or operate LLMs, should be aware of this high-severity vulnerability and take immediate action to protect their systems. This includes updating vLLM to version 0.24.0 or later, implementing input validation and sanitization for regular expression strings, and monitoring system performance and logs for signs of potential attacks. Additionally, security teams and vulnerability management teams should review the official advisory and CVE record to validate affected scope, severity, and vendor guidance.

Technical summary

The vLLM inference and serving engine for LLMs is vulnerable to a denial-of-service attack due to an issue with the structured_outputs.regex API parameter. Prior to version 0.24.0, the API parameter passes a user-supplied regular expression string directly to the grammar compiler backends with no compilation timeout. This allows an attacker to cause exponential state-space expansion, hanging an inference worker indefinitely and denying service. The issue is fixed in version 0.24.0 of vLLM.

Defensive priority

High priority should be given to updating vLLM to version 0.24.0 or later to prevent this denial-of-service vulnerability.

Recommended defensive actions

• Update vLLM to version 0.24.0 or later
• Implement input validation and sanitization for regular expression strings
• Monitor system performance and logs for signs of potential attacks
• Consider implementing rate limiting or other protective measures
• 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

Evidence notes

The CVE record and NVD detail provide information on the vulnerability and its impact. The vendor has released a patch for the issue, which is included in version 0.24.0 of vLLM. To verify the vulnerability, defenders should review the official advisory and CVE record, and check for affected product deployments in managed environments. The structured_outputs.regex API parameter passes a user-supplied regular expression string directly to the grammar compiler backends with no compilation timeout, allowing an attacker to cause exponential state-space expansion. This issue can be mitigated by updating vLLM to version 0.24.0 or later, implementing input validation and sanitization for regular expression strings, and monitoring system performance and logs for signs of potential attacks.

Official resources